RFC 2227






Network Working Group                                           J. Mogul
Request for Comments: 2227                                        DECWRL
Category: Standards Track                                       P. Leach
                                                               Microsoft
                                                            October 1997


            Simple Hit-Metering and Usage-Limiting for HTTP

Status of this Memo



   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice



   Copyright (C) The Internet Society (1997).  All Rights Reserved.

ABSTRACT



   This document proposes a simple extension to HTTP, using a new
   "Meter" header, which permits a limited form of demographic
   information (colloquially called "hit-counts") to be reported by
   caches to origin servers, in a more efficient manner than the
   "cache-busting" techniques currently used.  It also permits an origin
   server to control the number of times a cache uses a cached response,
   and outlines a technique that origin servers can use to capture
   referral information without "cache-busting."

TABLE OF CONTENTS

   1 Introduction                                                      2
        1.1 Goals, non-goals, and limitations                          3
        1.2 Brief summary of the design                                4
        1.3 Terminology                                                5
   2 Overview                                                          5
        2.1 Discussion                                                 7
   3 Design concepts                                                   8
        3.1 Implementation of the "metering subtree"                   8
        3.2 Format of the Meter header                                10
        3.3 Negotiation of hit-metering and usage-limiting            10
        3.4 Transmission of usage reports                             14
        3.5 When to send usage reports                                15
        3.6 Subdivision of usage-limits                               16




Mogul & Leach               Standards Track                     [Page 1]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   4 Analysis                                                         17
        4.1 Approximation accuracy for counting users                 18
        4.2 What about "Network Computers"?                           19
        4.3 Critical-path delay analysis                              19
   5 Specification                                                    20
        5.1 Specification of Meter header and directives              20
        5.2 Abbreviations for Meter directives                        23
        5.3 Counting rules                                            24
             5.3.1 Counting rules for hit-metering                    24
             5.3.2 Counting rules for usage-limiting                  25
             5.3.3 Equivalent algorithms are allowed                  26
        5.4 Counting rules: interaction with Range requests           27
        5.5 Implementation by non-caching proxies                     27
        5.6 Implementation by cooperating caches                      28
   6 Examples                                                         28
        6.1 Example of a complete set of exchanges                    28
        6.2 Protecting against HTTP/1.0 proxies                       30
        6.3 More elaborate examples                                   30
   7 Interactions with content negotiation                            31
        7.1 Treatment of responses carrying a Vary header             31
        7.2 Interaction with Transparent Content Negotiation          32
   8 A Note on Capturing Referrals                                    32
   9 Alternative proposals                                            33
   10 Security Considerations                                         34
   11 Acknowledgments                                                 35
   12 References                                                      35
   13 Authors' Addresses                                              36
   14 Full Copyright Statement                                        37

1 Introduction



   For a variety of reasons, content providers want to be able to
   collect information on the frequency with which their content is
   accessed. This desire leads to some of the "cache-busting" done by
   existing servers.  ("Cache-busting" is the use by servers of
   techniques intended to prevent caching of responses; it is unknown
   exactly how common this is.)  This kind of cache-busting is done not
   for the purpose of maintaining transparency or security properties,
   but simply to collect demographic information.  Some cache-busting is
   also done to provide different advertising images to appear on the
   same page (i.e., each retrieval of the page sees a different ad).

   This proposal supports a model similar to that of publishers of
   hard-copy publications: such publishers (try to) report to their
   advertisers how many people read an issue of a publication at least
   once; they don't (try to) report how many times a reader re-reads an
   issue. They do this by counting copies published, and then try to
   estimate, for their publication, on average how many people read a



Mogul & Leach               Standards Track                     [Page 2]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   single copy at least once. The key point is that the results aren't
   exact, but are still useful. Another model is that of coding
   inquiries in such a way that the advertiser can tell which
   publication produced the inquiry.

1.1 Goals, non-goals, and limitations



   HTTP/1.1 already allows origin servers to prevent caching of
   responses, and evidence exists [9] that at least some of the time,
   this is being done for the sole purpose of collecting counts of the
   number of accesses of specific pages.  Some of this evidence is
   inferred from the study of proxy traces; some is based on explicit
   statements of the intention of the operators of Web servers.
   Information collected this way might or might not be of actual use to
   the people who collect it; the fact is that they want to collect it,
   or already do so.

   The goal of this proposal is to provide an optional performance
   optimization for this use of HTTP/1.1.

   This specification is:

      - Optional: no server or proxy is required to implement it.

      - Proxy-centered: there is no involvement on the part of
        end-client implementations.

      - Solely a performance optimization: it provides no
        information or functionality that is not already available
        in HTTP/1.1.  The intent is to improve performance overall,
        and reduce latency for almost all interactions; latency
        might be increased for a small fraction of HTTP
        interactions.

      - Best-efforts: it does not guarantee the accuracy of the
        reported information, although it does provide accurate
        results in the absence of persistent network failures or
        host crashes.

      - Neutral with respect to privacy: it reveals to servers no
        information about clients that is not already available
        through the existing features of HTTP/1.1.

   The goals of this specification do not include:

      - Solving the entire problem of efficiently obtaining
        extensive information about requests made via proxies.




Mogul & Leach               Standards Track                     [Page 3]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      - Improving the protection of user privacy (although our
        proposal may reduce the transfer of user-specific
        information to servers, it does not prevent it).

      - Preventing or encouraging the use of log-exchange
        mechanisms.

      - Avoiding all forms of "cache-busting", or even all
        cache-busting done for gathering counts.

   This design has certain potential limitations:

      - If it is not deployed widely in both proxies and servers,
        it will provide little benefit.

      - It may, by partially solving the hit-counting problem,
        reduce the pressure to adopt more complete solutions, if
        any become available.

      - Even if widely deployed, it might not be widely used, and
        so might not significantly improve performance.

   These potential limitations might not be problems in actual practice.

1.2 Brief summary of the design



   This section is included for people not wishing to read the entire
   document; it is not a specification for the proposed design, and
   over-simplifies many aspects of the design.

   The goal of this design is to eliminate the need for origin servers
   to use "cache-busting" techniques, when this is done just for the
   purpose of counting the number of users of a resource.  (Cache-
   busting includes techniques such as setting immediate Expiration
   dates, or sending "Cache-control:  private" in each response.)

   The design adds a new "Meter" header to HTTP; the header is always
   protected by the "Connection" header, and so is always hop-by-hop.
   This mechanism allows the construction of a "metering subtree", which
   is a connected subtree of proxies, rooted at an origin server.  Only
   those proxies that explicitly volunteer to join in the metering
   subtree for a resource participate in hit-metering, but those proxies
   that do volunteer are required to make their best effort to provide
   accurate counts.  When a hit-metered response is forwarded outside of
   the metering subtree, the forwarding proxy adds "Cache-control: s-
   maxage=0", so that other proxies (outside the metering subtree) are
   forced to forward all requests to a server in the metering subtree.




Mogul & Leach               Standards Track                     [Page 4]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      NOTE: the HTTP/1.1 specification does not currently define a "s-
      maxage" Cache-control directive.  The HTTP working group has
      decided to add such a directive to the next revision of the
      HTTP/1.1 specification [7].

   The Meter header carries zero or more directives, similar to the way
   that the Cache-control header carries directives.  Proxies may use
   certain Meter directives to volunteer to do hit-metering for a
   resource.  If a proxy does volunteer, the server may use certain
   directives to require that a response be hit-metered.  Finally,
   proxies use a "count" Meter directive to report the accumulated hit
   counts.

   The Meter mechanism can also be used by a server to limit the number
   of uses that a cache may make of a cached response, before
   revalidating it.

   The full specification includes complete rules for counting "uses" of
   a response (e.g., non-conditional GETs) and "reuses" (conditional
   GETs).  These rules ensure that the results are entirely consistent
   in all cases, except when systems or networks fail.

1.3 Terminology



   This document uses terms defined and explained in the HTTP/1.1
   specification [4], including "origin server," "resource," "hop-by-
   hop," "unconditional GET," and "conditional GET."  The reader is
   expected to be familiar with the HTTP/1.1 specification and its
   terminology.

   The key words "MUST", "MUST NOT", "REQUIRED", "SHOULD", SHOULD NOT",
   "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be
   interpreted as described in RFC 2119 [1].

2 Overview



   The design described in this document introduces several new features
   to HTTP:

      - Hit-metering: allows an origin server to obtain reasonably
        accurate counts of the number of clients using a resource
        instance via a proxy cache, or a hierarchy of proxy caches.

      - Usage-limiting: allows an origin server to control the
        number of times a cached response may be used by a proxy
        cache, or a hierarchy of proxy caches, before revalidation
        with the origin server.




Mogul & Leach               Standards Track                     [Page 5]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   These new non-mandatory features require minimal new protocol
   support, no change in protocol version, relatively little overhead in
   message headers.  The design adds no additional network round-trips
   in any critical path that directly affects user-perceived latency
   (see section 4.3 for an analysis).

   The primary goal of hit-metering and usage-limiting is to obviate the
   need for an origin server to send "Cache-control: s-maxage=0" with
   responses for resources whose value is not likely to change
   immediately.  In other words, in cases where the only reason for
   contacting the origin server on every request that might otherwise be
   satisfied by a proxy cache entry is to allow the server to collect
   demographic information or to control the number of times a cache
   entry is used, the extension proposed here will avoid a significant
   amount of unnecessary network traffic and latency.

   This design introduces one new "Meter" header, which is used both in
   HTTP request messages and HTTP response messages.  The Meter header
   is used to transmit a number of directives and reports.  In
   particular, all negotiation of the use of hit-metering and usage
   limits is done using this header.  No other changes to the existing
   HTTP/1.1 specification [4] are proposed in this document.

   This design also introduces several new concepts:

      1. The concepts of a "use" of a cache entry, which is when a
         proxy returns its entity-body in response to a conditional
         or non-conditional request, and the "reuse" of a cache
         entry, which is when a proxy returns a 304 (Not Modified)
         response to a conditional request which is satisfied by
         that cache entry.

      2. The concept of a hit-metered resource, for which proxy
         caches make a best-effort attempt to report accurate
         counts of uses and/or reuses to the origin server.

      3. The concept of a usage-limited resource, for which the
         origin server expects proxy caches to limit the number of
         uses and/or reuses.



   The new Meter directives and reports interact to allow proxy caches
   and servers to cooperate in the collection of demographic data.  The
   goal is a best-efforts approximation of the true number of uses
   and/or reuses, not a guaranteed exact count.







Mogul & Leach               Standards Track                     [Page 6]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   The new Meter directives also allow a server to bound the inaccuracy
   of a particular hit-count, by bounding the number of uses between
   reports.  It can also, for example, bound the number of times the
   same ad is shown because of caching.

   Section 7.1 describes a way to use server-driven content negotiation
   (the Vary header) that allows an HTTP origin server to flexibly
   separate requests into categories and count requests by category.
   Implementation of such a categorized hit counting is likely to be a
   very small modification to most implementations of Vary; some
   implementations may not require any modification at all.

2.1 Discussion



   Mapping this onto the publishing model, a proxy cache would increment
   the use-count for a cache entry once for each unconditional GET done
   for the entry, and once for each conditional GET that results in
   sending a copy of the entry to update a client's invalid cached copy.
   Conditional GETs that result in 304 (Not Modified) are not included
   in the use-count, because they do not result in a new user seeing the
   page, but instead signify a repeat view by a user that had seen it
   before.  However, 304 responses are counted in the reuse-count.
   HEADs are not counted at all, because their responses do not contain
   an entity-body.

   The Meter directives apply only to shared proxy caches, not to end-
   client (or other single-user) caches.  Single user caches should not
   use Meter, because their hits will be automatically counted as a
   result of the unconditional GET with which they first fetch the page,
   from either the origin-server or from a proxy cache.  Their
   subsequent conditional GETs do not result in a new user seeing the
   page.

   The mechanism specified here counts GETs; other methods either do not
   result in a page for the user to read, aren't cached, or are
   "written-through" and so can be directly counted by the origin
   server. (If, in the future, a "cachable POST" came into existence,
   whereby the entity-body in the POST request was used to select a
   cached response, then such POSTs would have to be treated just like
   GETs.)  The applicability of hit-metering to any new HTTP methods
   that might be defined in the future is currently unspecifiable.

   In the case of multiple caches along a path, a proxy cache does the
   obvious summation when it receives a use-count or reuse-count in a
   request from another cache.






Mogul & Leach               Standards Track                     [Page 7]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


3 Design concepts

   In order to allow the introduction of hit-metering and usage-limiting
   without requiring a protocol revision, and to ensure a reasonably
   close approximation of accurate counts, the negotiation of metering
   and usage-limiting is done hop-by-hop, not end-to-end.  If one
   considers the "tree" of proxies that receive, store, and forward a
   specific response, the intent of this design is that within some
   (possibly null) "metering subtree", rooted at the origin server, all
   proxies are using the hit-metering and/or usage-limiting requested by
   the origin server.

   Proxies at the leaves of this subtree will insert a "Cache-control:
   s-maxage=0" directive, which forces all other proxies (below this
   subtree) to check with a leaf of the metering subtree on every
   request.  However, it does not prevent them from storing and using
   the response, if the revalidation succeeds.

   No proxy is required to implement hit-metering or usage-limiting.
   However, any proxy that transmits the Meter header in a request MUST
   implement every unconditional requirement of this specification,
   without exception or amendment.

   This is a conservative design, which may sometimes fail to take
   advantage of hit-metering support in proxies outside the metering
   subtree.  However, it is likely that without the reliability offered
   by a conservative design, managers of origin servers with
   requirements for accurate approximations will not take advantage of
   any hit-metering proposal.

   The hit-metering/usage-limiting mechanism is designed to avoid any
   extra network round-trips in the critical path of any client request,
   and (as much as possible) to avoid excessively lengthening HTTP
   messages.

   The Meter header is used to transmit both negotiation information and
   numeric information.

   A formal specification for the Meter header appears in section 5; the
   following discussion uses an informal approach to improve clarity.

3.1 Implementation of the "metering subtree"



   The "metering subtree" approach is implemented in a simple,
   straightforward way by defining the new "Meter" header as one that
   MUST always be protected by a Connection header in every request or
   response.  I.e., if the Meter header is present in an HTTP message,
   that message:



Mogul & Leach               Standards Track                     [Page 8]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      1. MUST contain "Connection: meter", and MUST be handled
         according to the HTTP/1.1 specification of the Connection
         header.

      2. MUST NOT be sent in response to a request from a client
         whose version number is less than HTTP/1.1.

      3. MUST NOT be accepted from a client whose version number is
         less than HTTP/1.1.

   The reason for the latter two restrictions is to protect against
   proxies that might not properly implement the Connection header.
   Otherwise, a subtree that includes an HTTP/1.0 proxy might
   erroneously appear to be a metering subtree.

      Note: It appears that for the Connection header mechanism to
      function correctly, a system receiving an HTTP/1.0 (or lower-
      version) message that includes a Connection header must act as if
      this header, and all of the headers it protects, ought to have
      been removed from the message by an intermediate proxy.

      Although RFC2068 does not specifically require this behavior, it
      appears to be implied.  Otherwise, one could not depend on the
      stated property (section 14.10) that the protected options "MUST
      NOT
be communicated by proxies over further connections."  This
      should probably be clarified in a subsequent draft of the HTTP/1.1
      specification.

      This specification does not, in any way, propose a modification of
      the specification of the Connection header.

   From the point of view of an origin server, the proxies in a metering
   subtree work together to obey usage limits and to maintain accurate
   usage counts.  When an origin server specifies a usage limit, a proxy
   in the metering subtree may subdivide this limit among its children
   in the subtree as it sees fit.  Similarly, when a proxy in the
   subtree receives a usage report, it ensures that the hits represented
   by this report are summed properly and reported to the origin server.

   When a proxy forwards a hit-metered or usage-limited response to a
   client (proxy or end-client) not in the metering subtree, it MUST
   omit the Meter header, and it MUST add "Cache-control: s-maxage=0" to
   the response.








Mogul & Leach               Standards Track                     [Page 9]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


3.2 Format of the Meter header



   The Meter header is used to carry zero or more directives.  Multiple
   Meter headers may occur in an HTTP message, but according to the
   rules in section 4.2 of the HTTP/1.1 specification [4], they may be
   combined into a single header (and should be so combined, to reduce
   overhead).

   For example, the following sequence of Meter headers

       Meter: max-uses=3
       Meter: max-reuses=10
       Meter: do-report

   may be expressed as

       Meter: max-uses=3, max-reuses=10, do-report

3.3 Negotiation of hit-metering and usage-limiting



   An origin server that wants to collect hit counts for a resource, by
   simply forcing all requests to bypass any proxy caches, would respond
   to requests on the resource with "Cache-control: s-maxage=0".  (An
   origin server wishing to prevent HTTP/1.0 proxies from improperly
   caching the response could also send both "Expires: <now>", to
   prevent such caching, and "Cache-control: max-age=NNNN", to allow
   newer proxies to cache the response).

   The purpose of the Meter header is to obviate the need for "Cache-
   control: s-maxage=0" within a metering subtree.  Thus, any proxy may
   negotiate the use of hit-metering and/or usage-limiting with the
   next-hop server.  If this server is the origin server, or is already
   part of a metering subtree (rooted at the origin server), then it may
   complete the negotiation, thereby extending the metering subtree to
   include the new proxy.

   To start the negotiation, a proxy sends its request with one of the
   following Meter directives:

   will-report-and-limit
                   indicates that the proxy is willing and able to
                   return usage reports and will obey any usage-limits.

   wont-report     indicates that the proxy will obey usage-limits but
                   will not send usage reports.

   wont-limit      indicates that the proxy will not obey usage-limits
                   but will send usage reports.



Mogul & Leach               Standards Track                    [Page 10]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   A proxy willing to neither obey usage-limits nor send usage reports
   MUST NOT transmit a Meter header in the request.

   By definition, an empty Meter header:

       Meter:

   is equivalent to "Meter: will-report-and-limit", and so, by the
   definition of the Connection header (see section 14.10 of the
   HTTP/1.1 specification [4]), a request that contains

       Connection: Meter

   and no explicit Meter header is equivalent to a request that contains

       Connection: Meter
       Meter: will-report-and-limit

   This makes the default case more efficient.

   An origin server that is not interested in metering or usage-limiting
   the requested resource simply ignores the Meter header.

   If the server wants the proxy to do hit-metering and/or usage-
   limiting, its response should include one or more of the following
   Meter directives:

   For hit-metering:

   do-report       specifies that the proxy MUST send usage reports to
                   the server.

   dont-report     specifies that the proxy SHOULD NOT send usage
                   reports to the server.

   timeout=NNN     sets a metering timeout of NNN minutes, from the time
                   that this response was originated, for the reporting
                   of a hit-count.  If the proxy has a non-zero hit
                   count for this response when the timeout expires, it
                   MUST send a report to the server at or before that
                   time.  Implies "do-report".

   By definition, an empty Meter header in a response, or any Meter
   header that does not contain "dont-report", means "Meter: do-report";
   this makes a common case more efficient.






Mogul & Leach               Standards Track                    [Page 11]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      Note: an origin server using the metering timeout mechanism to
      bound the collection period over which hit-counts are obtained
      should adjust the timeout values in the responses it sends so that
      all responses generated within that period reach their metering
      timeouts at or before the end of that period.

      If the origin server simply sends a constant metering timeout T
      with each response for a resource, the reports that it receives
      will reflect activity over a period whose duration is between T
      and N*T (in the worst case), where N is the maximum depth of the
      metering subtree.

   For usage-limiting

   max-uses=NNN    sets an upper limit of NNN "uses" of the response,
                   not counting its immediate forwarding to the
                   requesting end-client, for all proxies in the
                   following subtree taken together.

   max-reuses=NNN  sets an upper limit of NNN "reuses" of the response
                   for all proxies in the following subtree taken
                   together.

   When a proxy has exhausted its allocation of "uses" or "reuses" for a
   cache entry, it MUST revalidate the cache entry (using a conditional
   request) before returning it in a response.  (The proxy SHOULD use
   this revalidation message to send a usage report, if one was
   requested and it is time to send it.  See sections 3.4 and 3.5.)

   These Meter response-directives apply only to the specific response
   that they are attached to.

      Note that the limit on "uses" set by the max-uses directive does
      not include the use of the response to satisfy the end-client
      request that caused the proxy's request to the server.  This
      counting rule supports the notion of a cache-initiated prefetch: a
      cache may issue a prefetch request, receive a max-uses=0 response,
      store that response, and then return that response (without
      revalidation) when a client makes an actual request for the
      resource.  However, each such response may be used at most once in
      this way, so the origin server maintains precise control over the
      number of actual uses.

   A server MUST NOT send a Meter header that would require a proxy to
   do something that it has not yet offered to do.  A proxy receiving a
   Meter response-directive asking the proxy to do something it did not
   volunteer to do SHOULD ignore that directive.




Mogul & Leach               Standards Track                    [Page 12]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   A proxy receiving a Meter header in a response MUST either obey it,
   or it MUST revalidate the corresponding cache entry on every access.
   (I.e., if it chooses not to obey the Meter header in a response, it
   MUST act as if the response included "Cache-control:  s-maxage=0".)

      Note: a proxy that has not sent the Meter header in a request for
      the given resource, and which has therefore not volunteered to
      honor Meter directives in a response, is not required to honor
      them.  If, in this situation, the server does send a Meter header
      in a response, this is a protocol error.  However, based on the
      robustness principle, the proxy may choose to interpret the Meter
      header as an implicit request to include "Cache-control: s-
      maxage=0" when it forwards the response, since this preserves the
      apparent intention of the server.

   A proxy that receives the Meter header in a request may ignore it
   only to the extent that this is consistent with its own duty to the
   next-hop server.  If the received Meter request header is
   inconsistent with that duty, or if no Meter request header is
   received and the response from the next-hop server requests any form
   of metering or limiting, then the proxy MUST add "Cache-control: s-
   maxage=0" to any response it forwards for that request.  (A proxy
   SHOULD NOT add or change the Expires header or max-age Cache-control
   directive.)

      For example, if proxy A receives a GET request from proxy B for
      URL X with "Connection: Meter", but proxy A's cached response for
      URL does not include any Meter directives, then proxy A may ignore
      the metering offer from proxy B.

      However, if proxy A has previously told the origin server "Meter:
      wont-limit" (implying will-report), and the cached response
      contains "Meter: do-report", and proxy B's request includes
      "Meter:  wont-report", then proxy B's offer is inconsistent with
      proxy A's duty to the origin server.  Therefore, in this case
      proxy A must add "Cache-control: s-maxage=0" when it returns the
      cached response to proxy B, and must not include a Meter header in
      this response.

   If a server does not want to use the Meter mechanism, and will not
   want to use it any time soon, it may send this directive:

   wont-ask        recommends that the proxy SHOULD NOT send any Meter
                   directives to this server.







Mogul & Leach               Standards Track                    [Page 13]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   The proxy SHOULD remember this fact for up to 24 hours.  This avoids
   virtually all unnecessary overheads for servers that do not wish to
   use or support the Meter header.  (This directive also implies
   "dont-report".)

3.4 Transmission of usage reports



   To transmit a usage report, a proxy sends the following Meter header
   in a request on the appropriate resource:

       Meter: count=NNN/MMM

   The first integer indicates the count of uses of the cache entry
   since the last report; the second integer indicates the count of
   reuses of the entry (see section 5.3 for rules on counting uses and
   reuses).  The transmission of a "count" directive in a request with
   no other Meter directive is also defined as an implicit transmission
   of a "will-report-and-limit" directive, to optimize the common case.
   (A proxy not willing to honor usage-limits would send "Meter:
   count=NNN/MMM, wont-limit" for its reports.)

   Note that when a proxy forwards a client's request and receives a
   response, the response that the proxy sends immediately to the
   requesting client is not counted as a "use".  I.e., the reported
   count is the number of times the cache entry was used, and not the
   number of times that the response was used.

   A proxy SHOULD NOT transmit "Meter: count=0/0", since this conveys no
   useful information.

   Usage reports MUST always be transmitted as part of a conditional
   request (such as a GET or HEAD), since the information in the
   conditional header (e.g., If-Modified-Since or If-None-Match) is
   required for the origin server to know which instance of a resource
   is being counted.  Proxys forwarding usage reports up the metering
   subtree MUST NOT change the contents of the conditional header, since
   otherwise this would result in incorrect counting.

   A usage report MUST NOT be transmitted as part of a forwarded request
   that includes multiple entity tags in an If-None-Match or If-Match
   header.

      Note: a proxy that offers its willingness to do hit-metering
      (report usage) must count both uses and reuses.  It is not
      possible to negotiate the reporting of one but not the other.






Mogul & Leach               Standards Track                    [Page 14]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


3.5 When to send usage reports



   A proxy that has offered to send usage reports to its parent in the
   metering subtree MUST send a usage report in each of these
   situations:

      1. When it forwards a conditional GET on the resource
         instance on behalf of one of its clients (if the GET is
         conditional on at most one entity-tag).

      2. When it forwards a conditional HEAD on the resource
         instance on behalf of one of its clients.

      3. When it must generate a conditional GET to satisfy a
         client request because the max-uses limit has been
         exceeded.

      4. Upon expiration of a metering timeout associated with a
         cache entry that has a non-zero hit-count.



      5. When it removes the corresponding non-zero hit-count entry
         from its storage for any reason including:



            - the proxy needs the storage space for another
              hit-count entry.

            - the proxy is not able to store more than one response
              per resource, and a request forwarded on behalf of a
              client has resulted in the receipt of a new response
              (one with a different entity-tag or last-modified
              time).

         Note that a cache might continue to store hit-count information
         even after having deleted the body of the response, so it is
         not necessary to report the hit-count when deleting the body;
         it is only necessary to report it if the proxy is about to
         "forget" a non-zero value.

   (Section 5.3 explains how hit-counts become zero or non-zero.)

   If the usage report is being sent because the proxy is about to
   remove the hit-count entry from its storage, or because of an expired
   metering timeout:

      - The proxy MUST send the report as part of a conditional
        HEAD request on the resource instance.





Mogul & Leach               Standards Track                    [Page 15]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      - The proxy is not required to retry the HEAD request if it
        fails (this is a best-efforts design).  To improve
        accuracy, however, the proxy SHOULD retry failed HEAD
        requests, subject to resource constraints.

      - The proxy is not required to serialize any other operation
        on the completion of this request.

      Note: proxy implementors are strongly encouraged to batch several
      HEAD-based reports to the same server, when possible, over a
      single persistent connection, to reduce network overhead as much
      as possible.  This may involve a non-naive algorithm for
      scheduling the deletion of hit-count entries.

   If the usage count is sent because of an arriving request that also
   carries a "count" directive, the proxy MUST combine its own (possibly
   zero) use and reuse counts with the arriving counts, and then attempt
   to forward the request.

   However, the proxy is not required to forward an arriving request
   with a "count" directive, provided that:

      - it can reply to the request using a cached response, in
        compliance with other requirements of the HTTP
        specification.

      - such a response does not exceed a max-uses limit.

      - it is not required to forward the request because of an
        expired metering timeout.

   If an arriving request carries a "count" directive, and the proxy no
   longer has a cache entry for the resource, the proxy MUST forward the
   "count" directive.  (This is, in any case, what a proxy without a
   suitable cache entry would normally do for any valid request it
   receives.)

3.6 Subdivision of usage-limits



   When an origin server specifies a usage limit, a proxy in the
   metering subtree may subdivide this limit among its children in the
   subtree as it sees fit.

   For example, consider the situation with two proxies P1 and P2, each
   of which uses proxy P3 as a way to reach origin server S. Imagine
   that S sends P3 a response with





Mogul & Leach               Standards Track                    [Page 16]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


       Meter: max-uses=10

   The proxies use that response to satisfy the current requesting end-
   client.  The max-uses directive in this example allows the
   combination of P1, P2, and P3 together to satisfy 10 additional end-
   client uses (unconditional GETs) for the resource.

   This specification does not constrain how P3 divides up that
   allocation among itself and the other proxies.  For example, P3 could
   retain all of max-use allocation for itself.  In that case, it would
   forward the response to P1 and/or P2 with

       Meter: max-uses=0

   P3 might also divide the allocation equally among P1 and P2,
   retaining none for itself (which may be the right choice if P3 has
   few or no other clients).  In this case, it could send

       Meter: max-uses=5

   to the proxy (P1 or P2) that made the initial request, and then
   record in some internal data structure that it "owes" the other proxy
   the rest of the allocation.

   Note that this freedom to choose the max-uses value applies to the
   origin server, as well.  There is no requirement that an origin
   server send the same max-uses value to all caches.  For example, it
   might make sense to send "max-uses=2" the first time one hears from a
   cache, and then double the value (up to some maximum limit) each time
   one gets a "use-count" from that cache.  The idea is that the faster
   a cache is using up its max-use quota, the more likely it will be to
   report a use-count value before removing the cache entry.  Also, high
   and frequent use-counts imply a corresponding high efficiency benefit
   from allowing caching.

   Again, the details of such heuristics would be outside the scope of
   this specification.

4 Analysis

   This section includes informal analyses of several aspects of hit-
   metering:

      1. the accuracy of results when applied to counting users
         (section 4.1).

      2. the problem of counting users whose browsers do not
         include caches, such as Network Computers (section 4.2).



Mogul & Leach               Standards Track                    [Page 17]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      3. delays imposed on "critical paths" for HTTP operations
         (section 4.3).

4.1 Approximation accuracy for counting users



   For many (but not all) service operators, the single most important
   aspect of the request stream is the number of distinct users who have
   retrieved a particular entity within a given period (e.g., during a
   given day).  The hit-metering mechanism is designed to provide an
   origin server with an approximation of the number of users that
   reference a given resource.  The intent of the design is that the
   precision of this approximation is consistent with the goals of
   simplicity and optional implementation.

   Almost all Web users use client software that maintains local caches,
   and the state of the art of local-caching technology is quite
   effective.  (Section 4.2 discusses the case where end-client caches
   are small or non-existent.)  Therefore, assuming an effective and
   persistent end-client cache, each individual user who retrieves an
   entity does exactly one GET request that results in a 200 or 203
   response, or a 206 response that includes the first byte of the
   entity. If a proxy cache maintains and reports an accurate use-count
   of such retrievals, then its reported use-count will closely
   approximate the number of distinct users who have retrieved the
   entity.

   There are some circumstances under which this approximation can break
   down.  For example, if an entity stays in a proxy cache for much
   longer than it persists in the typical client cache, and users often
   re-reference the entity, then this scheme will tend to over-count the
   number of users. Or, if the cache-management policy implemented in
   typical client caches is biased against retaining certain kinds of
   frequently re-referenced entities (such as very large images), the
   use-counts reported will tend to overestimate the user-counts for
   such entities.

   Browser log analysis has shown that when a user revisits a resource,
   this is almost always done very soon after the previous visit, almost
   always with fewer than eight intervening references [11].  Although
   this result might not apply universally, it implies that almost all
   reuses will hit in the end-client cache, and will not be seen as
   unconditional GETs by a proxy cache.

   The existing (HTTP/1.0) "cache-busting" mechanisms for counting
   distinct users will certainly overestimate the number of users behind
   a proxy, since it provides no reliable way to distinguish between a
   user's initial request and subsequent repeat requests that might have
   been conditional GETs, had not cache-busting been employed.  The



Mogul & Leach               Standards Track                    [Page 18]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   "Cache-control: s-maxage=0" feature of HTTP/1.1 does allow the
   separation of use-counts and reuse-counts, provided that no HTTP/1.0
   proxy caches intervene.

   Note that if there is doubt about the validity of the results of
   hit-metering a given set of resources, the server can employ cache-
   busting techniques for short periods, to establish a baseline for
   validating the hit-metering results.  Various approaches to this
   problem are discussed in a paper by James Pitkow [9].

4.2 What about "Network Computers"?



   The analysis in section 4.1 assumed that "almost all Web users" have
   client caches.  If the Network Computers (NC) model becomes popular,
   however, then this assumption may be faulty: most proposed NCs have
   no disk storage, and relatively little RAM.  Many Personal Digital
   Assistants (PDAs), which sometimes have network access, have similar
   constraints.  Such client systems may do little or no caching of HTTP
   responses.  This means that a single user might well generate many
   unconditional GETs that yield the same response from a proxy cache.

   First note that the hit-metering design in this document, even with
   such clients, provides an approximation no worse than available with
   unmodified HTTP/1.1: the counts that a proxy would return to an
   origin server would represent exactly the number of requests that the
   proxy would forward to the server, if the server simply specifies
   "Cache-control:  s-maxage=0".

   However, it may be possible to improve the accuracy of these hit-
   counts by use of some heuristics at the proxy.  For example, the
   proxy might note the IP address of the client, and count only one GET
   per client address per response.  This is not perfect: for example,
   it fails to distinguish between NCs and certain other kinds of hosts.
   The proxy might also use the heuristic that only those clients that
   never send a conditional GET should be treated this way, although we
   are not at all certain that NCs will never send conditional GETs.

   Since the solution to this problem appears to require heuristics
   based on the actual behavior of NCs (or perhaps a new HTTP protocol
   feature that allows unambiguous detection of cacheless clients), it
   appears to be premature to specify a solution.

4.3 Critical-path delay analysis



   In systems (such as the Web) where latency is at issue, there is
   usually a tree of steps which depend on one another, in such a way
   that the final result cannot be accomplished until all of its
   predecessors have been.  Since the tree structure admits some



Mogul & Leach               Standards Track                    [Page 19]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   parallelism, it is not necessary to add up the timings for each step
   to discover the latency for the entire process.  But any single path
   through this dependency tree cannot be parallelized, and the longest
   such path is the one whose length (in units of seconds) determines
   the overall latency.  This is the "critical path", because no matter
   how much shorter one makes any other path, that cannot change the
   overall latency for the final result.

   If one views the final result, for a Web request, as rendering a page
   at a browser, or otherwise acting on the result of a request, clearly
   some network round trips (e.g., exchanging TCP SYN packets if the
   connection doesn't already exist) are on the critical path.  This
   hit-metering design does add some round-trips for reporting non-zero
   counts when a cache entry is removed, but, by design, these are off
   any critical path:  they may be done in parallel with any other
   operation, and require only "best efforts", so a proxy does not have
   to serialize other operations with their success or failure.

   Clearly, anything that changes network utilization (either increasing
   or decreasing it) can indirectly affect user-perceived latency.  Our
   expectation is that hit-metering, on average, will reduce loading and
   so even its indirect effects should not add network round-trips in
   any critical path.  But there might be a few specific instances where
   the added non-critical-path operations (specifically, usage reports
   upon cache-entry removal) delay an operation on a critical path.
   This is an unavoidable problem in datagram networks.

5 Specification

5.1 Specification of Meter header and directives



   The Meter general-header field is used to:

      - Negotiate the use of hit-metering and usage-limiting among
        origin servers and proxy caches.

      - Report use counts and reuse counts.

   Implementation of the Meter header is optional for both proxies and
   origin servers.  However, any proxy that transmits the Meter header
   in a request MUST implement every requirement of this specification,
   without exception or amendment.

   The Meter header MUST always be protected by a Connection header.  A
   proxy that does not implement the Meter header MUST NOT pass it
   through to another system (see section 5.5 for how a non-caching
   proxy may comply with this specification).  If a Meter header is




Mogul & Leach               Standards Track                    [Page 20]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   received in a message whose version is less than HTTP/1.1, it MUST be
   ignored (because it has clearly flowed through a proxy that does not
   implement Meter).

   A proxy that has received a response with a version less than
   HTTP/1.1, and therefore from a server (or another proxy) that does
   not implement the Meter header, SHOULD NOT send Meter request
   directives to that server, because these would simply waste
   bandwidth.  This recommendation does not apply if the proxy is
   currently hit-metering or usage-limiting any responses from that
   server.  If the proxy receives a HTTP/1.1 or higher response from
   such a server, it should cease its suppression of the Meter
   directives.

   All proxies sending the Meter header MUST adhere to the "metering
   subtree" design described in section 3.

       Meter = "Meter" ":" 0#meter-directive

       meter-directive = meter-request-directive
                       | meter-response-directive
                       | meter-report-directive

       meter-request-directive =
                         "will-report-and-limit"
                       | "wont-report"
                       | "wont-limit"

       meter-report-directive =
                       | "count" "=" 1*DIGIT "/" 1*DIGIT

       meter-response-directive =
                         "max-uses" "=" 1*DIGIT
                       | "max-reuses" "=" 1*DIGIT
                       | "do-report"
                       | "dont-report"
                       | "timeout" "=" 1*DIGIT
                       | "wont-ask"

   A meter-request-directive or meter-report-directive may only appear
   in an HTTP request message.  A meter-response-directive may only
   appear in an HTTP response directive.

   An empty Meter header in a request means "Meter: will-report-and-
   limit".  An empty Meter header in a response, or any other response
   including one or more Meter headers without the "dont-report" or
   "wont-ask" directive, implies "Meter:  do-report".




Mogul & Leach               Standards Track                    [Page 21]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   The meaning of the meter-request-directives are as follows:

   will-report-and-limit
                   indicates that the proxy is willing and able to
                   return usage reports and will obey any usage-limits.

   wont-report     indicates that the proxy will obey usage-limits but
                   will not send usage reports.

   wont-limit      indicates that the proxy will not obey usage-limits
                   but will send usage reports.

   A proxy willing neither to obey usage-limits nor to send usage
   reports MUST NOT transmit a Meter header in the request.

   The meaning of the meter-report-directives are as follows:

   count "=" 1*DIGIT "/" 1*DIGIT
                   Both digit strings encode decimal integers.  The
                   first integer indicates the count of uses of the
                   cache entry since the last report; the second integer
                   indicates the count of reuses of the entry.

   Section 5.3 specifies the counting rules.

   The meaning of the meter-response-directives are as follows:

   max-uses "=" 1*DIGIT
                   sets an upper limit on the number of "uses" of the
                   response, not counting its immediate forwarding to
                   the requesting end-client, for all proxies in the
                   following subtree taken together.

   max-reuses "=" 1*DIGIT
                   sets an upper limit on the number of "reuses" of the
                   response for all proxies in the following subtree
                   taken together.

   do-report       specifies that the proxy MUST send usage reports to
                   the server.

   dont-report     specifies that the proxy SHOULD NOT send usage
                   reports to the server.

   timeout "=" 1*DIGIT
                   sets a metering timeout of the specified number of
                   minutes (not seconds) after the origination of this
                   response (as indicated by its "Date" header).  If the



Mogul & Leach               Standards Track                    [Page 22]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


                   proxy has a non-zero hit count for this response when
                   the timeout expires, it MUST send a report to the
                   server at or before that time.  Timeouts should be
                   implemented with an accuracy of plus or minus one
                   minute.  Implies "do-report".

   wont-ask        specifies that the proxy SHOULD NOT send any Meter
                   headers to the server.  The proxy should forget this
                   advice after a period of no more than 24 hours.

   Section 5.3 specifies the counting rules, and in particular specifies
   a somewhat non-obvious interpretation of the max-uses value.

5.2 Abbreviations for Meter directives



   To allow for the most efficient possible encoding of Meter headers,
   we define abbreviated forms of all Meter directives.  These are
   exactly semantically equivalent to their non-abbreviated
   counterparts.  All systems implementing the Meter header MUST
   implement both the abbreviated and non-abbreviated forms.
   Implementations SHOULD use the abbreviated forms in normal use.

   The abbreviated forms of Meter directive are shown below, with the
   corresponding non-abbreviated literals in the comments:

       Abb-Meter = "Meter" ":" 0#abb-meter-directive

       abb-meter-directive = abb-meter-request-directive
                       | abb-meter-response-directive
                       | abb-meter-report-directive

       abb-meter-request-directive =
                         "w"           ; "will-report-and-limit"
                       | "x"           ; "wont-report"
                       | "y"           ; "wont-limit"

       abb-meter-report-directive =
                       | "c" "=" 1*DIGIT "/" 1*DIGIT   ; "count"

       abb-meter-response-directive =
                         "u" "=" 1*DIGIT       ; "max-uses"
                       | "r" "=" 1*DIGIT       ; "max-reuses"
                       | "d"                   ; "do-report"
                       | "e"                   ; "dont-report"
                       | "t" "=" 1*DIGIT       ; "timeout"
                       | "n"                   ; "wont-ask"





Mogul & Leach               Standards Track                    [Page 23]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      Note: although the Abb-Meter BNF rule is defined separately from
      the Meter rule, one may freely mix abbreviated and non-abbreviated
      Meter directives in the same header.

5.3 Counting rules



      Note: please remember that hit-counts and usage-counts are
      associated with individual responses, not with resources.  A cache
      entry that, over its lifetime, holds more than one response is
      also not a "response", in this particular sense.

   Let R be a cached response, and V be the value of the Request-URI and
   selecting request-headers (if any, see section 14.43 of the HTTP/1.1
   specification [4]) that would select R if contained in a request.  We
   define a "use" of R as occurring when the proxy returns its stored
   copy of R in a response with any of the following status codes: a 200
   (OK) status; a 203 (Non-Authoritative Information) status; or a 206
   (Partial Content) status when the response contains byte #0 of the
   entity (see section 5.4 for a discussion of Range requests).

      Note: when a proxy forwards a client's request and receives a
      response, the response that the proxy sends immediately to the
      requesting client is not counted as a "use".  I.e., the reported
      count is the number of times the cache entry was used, and not the
      number of times that the response was used.

   We define a "reuse" of R as as occurring when the proxy responds to a
   request selecting R with a 304 (Not Modified) status, unless that
   request is a Range request that does not specify byte #0 of the
   entity.

5.3.1 Counting rules for hit-metering



   A proxy participating in hit-metering for a cache response R
   maintains two counters, CU and CR, associated with R. When a proxy
   first stores R in its cache, it sets both CU and CR to 0 (zero).
   When a subsequent client request results in a "use" of R, the proxy
   increments CU.  When a subsequent client request results in a "reuse"
   of R, the proxy increments CR.  When a subsequent client request
   selecting R (i.e., including V) includes a "count" Meter directive,
   the proxy increments CU and CR using the corresponding values in the
   directive.

   When the proxy sends a request selecting R (i.e., including V) to the
   inbound server, it includes a "count" Meter directive with the
   current CU and CR as the parameter values.  If this request was
   caused by the proxy's receipt of a request from a client, upon
   receipt of the server's response, the proxy sets CU and CR to the



Mogul & Leach               Standards Track                    [Page 24]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   number of uses and reuses, respectively, that may have occurred while
   the request was in progress.  (These numbers are likely, but not
   certain, to be zero.)  If the proxy's request was a final HEAD-based
   report, it need no longer maintain the CU and CR values, but it may
   also set them to the number of intervening uses and reuses and retain
   them.

5.3.2 Counting rules for usage-limiting



   A proxy participating in usage-limiting for a response R maintains
   either or both of two counters TU and TR, as appropriate, for that
   resource.  TU and TR are incremented in just the same way as CU and
   CR, respectively.  However, TU is zeroed only upon receipt of a
   "max-uses" Meter directive for that response (including the initial
   receipt).  Similarly, TR is zeroed only upon receipt of a "max-
   reuses" Meter directive for that response.

   A proxy participating in usage-limiting for a response R also stores
   values MU and/or MR associated with R. When it receives a response
   including only a max-uses value, it sets MU to that value and MR to
   infinity.  When it receives a response including only a max-reuses
   value, it sets MR to that value and MU to infinity.  When it receives
   a response including both max-reuses and max-reuses values, it sets
   MU and MR to those values, respectively.  When it receives a
   subsequent response including neither max-reuses nor max-reuses
   values, it sets both MU and MR to infinity.

   If a proxy participating in usage-limiting for a response R receives
   a request that would cause a "use" of R, and TU >= MU, it MUST
   forward the request to the server.  If it receives a request that
   would cause a "reuse" of R, and TR >= MR, it MUST forward the request
   to the server.  If (in either case) the proxy has already forwarded a
   previous request to the server and is waiting for the response, it
   should delay further handling of the new request until the response
   arrives (or times out); it SHOULD NOT have two revalidation requests
   pending at once that select the same response, unless these are Range
   requests selecting different subranges.

   There is a special case of this rule for the "max-uses" directive: if
   the proxy receives a response with "max-uses=0" and does not forward
   it to a requesting client, the proxy should set a flag PF associated
   with R. If R is true, then when a request arrives while if TU >= MU,
   if the PF flag is set, then the request need not be forwarded to the
   server (provided that this is not required by other caching rules).
   However, the PF flag MUST be cleared on any use of the response.






Mogul & Leach               Standards Track                    [Page 25]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


      Note: the "PF" flag is so named because this feature is useful
      only for caches that could issue a "prefetch" request before an
      actual client request for the response.  A proxy not implementing
      prefetching need not implement the PF flag.

5.3.3 Equivalent algorithms are allowed



   Any other algorithm that exhibits the same external behavior (i.e.,
   generates exactly the same requests from the proxy to the server) as
   the one in this section is explicitly allowed.

      Note: in most cases, TU will be equal to CU, and TR will be
      equal to CR.  The only two cases where they could differ are:

         1. The proxy issues a non-conditional request for the
            resource using V, while TU and/or TR are non-zero, and
            the server's response includes a new "max-uses" and/or
            "max-reuses" directive (thus zeroing TU and/or TR, but
            not CU and CR).

         2. The proxy issues a conditional request reporting the
            hit-counts (and thus zeroing CU and CR, but not TU or
            TR), but the server's response does not include a new
            "max-uses" and/or "max-reuses" directive.

      To solve the first case, the proxy has several implementation
      options

         - Always store TU and TR separately from CU and CR.

         - Create "shadow" copies of TU and TR when this situation
           arises (analogous to "copy on write").

         - Generate a HEAD-based usage report when the
           non-conditional request is sent (or when the
           "max-uses=0" is received), causing CU and CR to be
           zeroed (analogous in some ways to a "memory barrier"
           instruction).

      In the second case, the server implicitly has removed the
      usage-limit(s) on the response (by setting MU and/or MR to
      infinity), and so the fact that, say, TU is different from CU
      is not significant.

      Note: It may also be possible to eliminate the PF flag by
      sending extra HEAD-based usage-report requests, but we
      recommend against this; it is better to allocate an extra bit
      per entry than to transmit extra requests.



Mogul & Leach               Standards Track                    [Page 26]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


5.4 Counting rules: interaction with Range requests




   HTTP/1.1 allows a client to request sub-ranges of a resource.  A
   client might end up issuing several requests with the net effect of
   receiving one copy of the resource.  For uniformity of the results
   seen by origin servers, proxies need to observe a rule for counting
   these references, although it is not clear that one rule generates
   accurate results in every case.

   The rule established in this specification is that proxies count as a
   "use" or "reuse" only those Range requests that result in the return
   of byte #0 of the resource.  The rationale for this rule is that in
   almost every case, an end-client will retrieve the beginning of any
   resource that it references at all, and that it will seldom retrieve
   any portion more than once.  Therefore, this rule appears to meet the
   goal of a "best-efforts" approximation.

5.5 Implementation by non-caching proxies



   A non-caching proxy may participate in the metering subtree; this is
   strongly recommended.

   A non-caching proxy (HTTP/1.1 or higher) that participates in the
   metering subtree SHOULD forward Meter headers on both requests and
   responses, with the appropriate Connection headers.

   If a non-caching proxy forwards Meter headers, it MUST comply with
   these restrictions:

      1. If the proxy forwards Meter headers in responses, such a
         response MUST NOT be returned to any request except the
         one that elicited it.

      2. Once a non-caching proxy starts forwarding Meter headers,
         it should not arbitrarily stop forwarding them (or else
         reports may be lost).

   A proxy that caches some responses and not others, for whatever
   reason, may choose to implement the Meter header as a caching proxy
   for the responses that it caches, and as a non-caching proxy for the
   responses that it does not cache, as long as its external behavior
   with respect to any particularly response is fully consistent with
   this specification.







Mogul & Leach               Standards Track                    [Page 27]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


5.6 Implementation by cooperating caches



   Several HTTP cache implementations, most notably the Harvest/Squid
   cache [2], create cooperative arrangements between several caches.
   If such caches use a protocol other than HTTP to communicate between
   themselves, such as the Internet Cache Protocol (ICP) [12], and if
   they implement the Meter header, then they MUST act to ensure that
   their cooperation does not violate the intention of this
   specification.

   In particular, if one member of a group of cooperating caches agrees
   with a server to hit-meter a particular response, and then passes
   this response via a non-HTTP protocol to a second cache in the group,
   the caches MUST ensure that the server which requested the metering
   receives reports that appropriately account for any uses or resues
   made by the second cache.  Similarly, if the first cache agreed to
   usage-limit the response, the total number of uses by the group of
   caches MUST be limited to the agreed-upon number.

6 Examples



6.1 Example of a complete set of exchanges



   This example shows how the protocol is intended to be used most of
   the time: for hit-metering without usage-limiting.  Entity bodies are
   omitted.

   A client sends request to a proxy:

       GET http://foo.com/bar.html HTTP/1.1

   The proxy forwards request to the origin server:

       GET /bar.html HTTP/1.1
       Host: foo.com
       Connection: Meter

   thus offering (implicitly) "will-report-and-limit".

   The server responds to the proxy:

       HTTP/1.1 200 OK
       Date: Fri, 06 Dec 1996 18:44:29 GMT
       Cache-control: max-age=3600
       Connection: meter
       Etag: "abcde"





Mogul & Leach               Standards Track                    [Page 28]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   thus (implicitly) requiring "do-report" (but not requiring
   usage-limiting).

   The proxy responds to the client:

       HTTP/1.1 200 OK
       Date: Fri, 06 Dec 1996 18:44:29 GMT
       Etag: "abcde"
       Cache-control: max-age=3600, proxy-mustcheck
       Age: 1

   Since the proxy does not know if its client is an end-system, or a
   proxy that doesn't do metering, it adds the "proxy-mustcheck"
   directive.

   Another client soon asks for the resource:

       GET http://foo.com/bar.html HTTP/1.1

   and the proxy sends the same response as it sent to the other client,
   except (perhaps) for the Age value.

   After an hour has passed, a third client asks for the response:

       GET http://foo.com/bar.html HTTP/1.1

   But now the response's max-age has been exceeded, so the proxy
   revalidates the response with the origin server:

       GET /bar.html HTTP/1.1
       If-None-Match: "abcde"
       Host: foo.com
       Connection: Meter
       Meter: count=1/0

   thus simultaneously fulfilling its duties to validate the response
   and to report the one "use" that wasn't forwarded.

   The origin server responds:

       HTTP/1.1 304 Not Modified
       Date: Fri, 06 Dec 1996 19:44:29 GMT
       Cache-control: max-age=3600
       Etag: "abcde"

   so the proxy can use the original response to reply to the new
   client; the proxy also zeros the use-count it associates with that
   response.



Mogul & Leach               Standards Track                    [Page 29]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   Another client soon asks for the resource:

       GET http://foo.com/bar.html HTTP/1.1

   and the proxy sends the appropriate response.

   After another few hours, the proxy decides to remove the cache entry.
   When it does so, it sends to the origin server:

       HEAD /bar.html HTTP/1.1
       If-None-Match: "abcde"
       Host: foo.com
       Connection: Meter
       Meter: count=1/0

   reporting that one more use of the response was satisfied from the
   cache.

6.2 Protecting against HTTP/1.0 proxies



   An origin server that does not want HTTP/1.0 caches to store the
   response at all, and is willing to have HTTP/1.0 end-system clients
   generate excess GETs (which will be forwarded by HTTP/1.0 proxies)
   could send this for its reply:

       HTTP/1.1 200 OK
       Cache-control: max-age=3600
       Connection: meter
       Etag: "abcde"
       Expires: Sun, 06 Nov 1994 08:49:37 GMT

   HTTP/1.0 caches will see the ancient Expires header, but HTTP/1.1
   caches will see the max-age directive and will ignore Expires.

      Note: although most major HTTP/1.0 proxy implementations observe
      the Expires header, it is possible that some are in use that do
      not.  Use of the Expires header to prevent caching by HTTP/1.0
      proxies might not be entirely reliable.

6.3 More elaborate examples



   Here is a request from a proxy that is willing to hit-meter but is
   not willing to usage-limit:

       GET /bar.html HTTP/1.1
       Host: foo.com
       Connection: Meter
       Meter: wont-limit



Mogul & Leach               Standards Track                    [Page 30]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   Here is a response from an origin server that does not want hit
   counting, but does want "uses" limited to 3, and "reuses" limited to
   6:

       HTTP/1.1 200 OK
       Cache-control: max-age=3600
       Connection: meter
       Etag: "abcde"
       Expires: Sun, 06 Nov 1994 08:49:37 GMT
       Meter: max-uses=3, max-reuses=6, dont-report

   Here is the same example with abbreviated Meter directive names:

       HTTP/1.1 200 OK
       Cache-control: max-age=3600
       Connection: meter
       Etag: "abcde"
       Expires: Sun, 06 Nov 1994 08:49:37 GMT
       Meter:u=3,r=6,e

7 Interactions with content negotiation



   This section describes two aspects of the interaction between hit-
   metering and "content-negotiated" resources:

      1. treatment of responses carrying a Vary header (section
         7.1).

      2. treatment of responses that use the proposed Transparent
         Content Negotiation mechanism (section 7.2).

7.1 Treatment of responses carrying a Vary header



   Separate counts should be kept for each combination of the headers
   named in the Vary header for the Request-URI (what [4] calls "the
   selecting request-headers"), even if they map to the same entity-tag.
   This rule has the effect of counting hits on each variant, if there
   are multiple variants of a page available.

      Note: This interaction between Vary and the hit-counting
      directives allows the origin server a lot of flexibility in
      specifying how hits should be counted.  In essence, the origin
      server uses the Vary mechanism to divide the requests for a
      resource into arbitrary categories, based on the request- headers.
      (We will call these categories "request-patterns".) Since a proxy
      keeps its hit-counts for each request-pattern, rather than for
      each resource, the origin server can obtain separate statistics
      for many aspects of an HTTP request.



Mogul & Leach               Standards Track                    [Page 31]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   For example, if a page varied based on the value of the User-Agent
   header in the requests, then hit counts would be kept for each
   different flavor of browser. But it is in fact more general than
   that; because multiple header combinations can map to the same
   variant, it also enables the origin server to count the number of
   times (e.g.) the Swahili version of a page was requested, even though
   it is only available in English.

   If a proxy does not support the Vary mechanism, then [4] says that it
   MUST NOT cache any response that carries a Vary header, and hence
   need not implement any aspect of this hit-counting or usage-limiting
   design for varying resources.

       Note: this also implies that if a proxy supports the Vary
       mechanism but is not willing to maintain independent hit-counts
       for each variant response in its cache, then it must follow at
       least one of these rules:

          1. It must not use the Meter header in a request to offer
             to hit-meter or usage-limit responses.

          2. If it does offer to hit-meter or usage-limit responses,
             and then receives a response that includes both a Vary
             header and a Meter header with a directive that it
             cannot satisfy, then the proxy must not cache the
             response.

       In other words, a proxy is allowed to partially implement the
       Vary mechanism with respect to hit-metering, as long as this has
       no externally visible effect on its ability to comply with the
       Meter specification.

   This approach works for counting almost any aspect of the request
   stream, without embedding any specific list of countable aspects in
   the specification or proxy implementation.

7.2 Interaction with Transparent Content Negotiation



   [A description of the interaction between this design and the
   proposed Transparent Content Negotiation (TCN) design [6] will be
   made available in a later document.]

8 A Note on Capturing Referrals



   It is alleged that some advertisers want to pay content providers,
   not by the "hit", but by the "nibble" -- the number of people who
   actually click on the ad to get more information.




Mogul & Leach               Standards Track                    [Page 32]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   Now, HTTP already has a mechanism for doing this: the "Referer"
   header. However, perhaps it ought to be disabled for privacy reasons
   -- according the HTTP/1.1 spec:

       "Because the source of the link may be private information or may
       reveal an otherwise private information source, it is strongly
       recommended that the user be able to select whether or not the
       Referer field is sent."

   However, in the case of ads, the source of the link actually wants to
   let the referred-to page know where the reference came from.

   This does not require the addition of any extra mechanism, but rather
   can use schemes that embed the referrer in the URI in a manner
   similar to this:

          http://www.blah.com/ad-reference?from=site1

   Such a URI should point to a resource (perhaps a CGI script) which
   returns a 302 redirect to the real page

          http://www.blah.com/ad-reference.html

   Proxies which do not cache 302s will cause one hit on the redirection
   page per use, but the real page will get cached. Proxies which do
   cache 302s and report hits on the cached 302s will behave optimally.

   This approach has the advantage that it works whether or not the
   end-client has disabled the use of Referer.  Combined with the rest
   of the hit-metering proposal in this design, this approach allows,
   for example, an advertiser to know how often a reference to an
   advertisement was made from a particular page.

9 Alternative proposals



   There might be a number of other ways of gathering demographic and
   usage information; other mechanisms might respond to a different set
   of needs than this proposal does.  This proposal certainly does not
   preclude the proposal or deployment of other such mechanisms, and
   many of them may be complementary to and compatible with the
   mechanism proposed here.

   There has been some speculation that statistical sampling methods
   might be used to gather reasonably accurate data.  One such proposal
   is to manipulate cache expiration times so that selected resources
   are uncachable for carefully chosen periods, allowing servers to
   accurately count accesses during those periods.  The hit-metering
   mechanism proposed here is entirely complementary to that approach,



Mogul & Leach               Standards Track                    [Page 33]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   since it could be used to reduce the cost of gathering those counts.
   James Pitkow has written a paper comparing an earlier draft of this
   hit-metering proposal with sampling approaches [9].

   Phillip Hallam-Baker has proposed using a log-exchange protocol [5],
   by which a server could request a proxy's logs by making an HTTP
   request to the proxy.  This proposal asserts that it is "believed to
   operate correctly in configurations involving multiple proxies", but
   it is not clear that this is true if an outer proxy is used as a
   (one-way) firewall.  The proposal also leaves a number of open
   issues, such as how an origin server can be sure that all of the
   proxies in the request subtree actually support log-exchange.  It is
   also not clear how this proposal couples a proxy's support of log-
   exchange to a server's permission to cache a response.

   For general background on the topic of Web measurement standards, see
   the discussion by Thomas P. Novak and Donna L. Hoffman [8].  Also see
   the "Privacy and Demographics Overview" page maintained by by the
   World Wide Web Consortium [10], which includes a pointer to some
   tentative proposals for gathering consumer demographics (not just
   counting references) [3].

10 Security Considerations



   Which outbound clients should a server (proxy or origin) trust to
   report hit counts?  A malicious proxy could easily report a large
   number of hits on some page, and thus perhaps cause a large payment
   to a content provider from an advertiser.  To help avoid this
   possibility, a proxy may choose to only relay usage counts received
   from its outbound proxies to its inbound servers when the proxies
   have authenticated themselves using Proxy-Authorization and/or they
   are on a list of approved proxies.

   It is not possible to enforce usage limits if a proxy is willing to
   cheat (i.e., it offers to limit usage but then ignores a server's
   Meter directive).

   Regarding privacy:  it appears that the design in this document does
   not reveal any more information about individual users than would
   already be revealed by implementation of the existing HTTP/1.1
   support for "Cache-control: max-age=0, proxy-revalidate" or "Cache-
   control: s-maxage=0".  It may, in fact, help to conceal certain
   aspects of the organizational structure on the outbound side of a
   proxy.  In any case, the conflict between user requirements for
   anonymity and origin server requirements for demographic information
   cannot be resolved by purely technical means.





Mogul & Leach               Standards Track                    [Page 34]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


11 Acknowledgments



   We gratefully acknowledge the constructive comments received from
   Anselm Baird-Smith, Ted Hardie, Koen Holtman (who suggested the
   technique described in section 8), Dave Kristol, Ari Luotonen,
   Patrick R. McManus, Ingrid Melve, and James Pitkow.

12 References



   1.  Bradner, S.,  "Key words for use in RFCs to Indicate Requirement
       Levels", BCP 14, RFC 2119, March 1997.

   2.  Anwat Chankhunthod, Peter B. Danzig, Chuck Neerdaels, Michael
       F. Schwartz, and Kurt J. Worrell.  A Hierarchical Internet Object
       Cache.  Proc. 1996 USENIX Technical Conf., San Diego, January,
       1996, pp. 153-163.

   3.  Daniel W. Connolly.  Proposals for Gathering Consumer
       Demographics.
       http://www.w3.org/pub/WWW/Demographics/Proposals.html.

   4.  Fielding, R., Gettys, J., Mogul, J., Nielsen, H. and T.
       Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1," RFC 2068,
       January, 1997.

   5.  Phillip M. Hallam-Baker.  Notification for Proxy Caches.  W3C
       Working Draft WD-proxy-960221, World Wide Web Consortium,
       February, 1996. http://www.w3.org/pub/WWW/TR/WD-proxy.html.

   6.  Holtman, K., and A. Mutz, "Transparent Content Negotiation in
       HTTP", Work in Progress.

   7.  Mogul, J., "Forcing HTTP/1.1 proxies to revalidate responses",
       Work in Progress.

   8.  Thomas P. Novak and Donna L. Hoffman.  New Metrics for New Media:
       Toward the Development of Web Measurement Standards.  This is a
       draft paper, currently available at http://
       www2000.ogsm.vanderbilt.edu/novak/web.standards/webstand.html.
       Cited by permission of the author; do not quote or cite without
       permission.

   9.  James Pitkow.  In search of reliable usage data on the WWW.
       Proc. Sixth International World Wide Web Conference, Santa Clara,
       CA, April, 1997.






Mogul & Leach               Standards Track                    [Page 35]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


   10. Joseph Reagle, Rohit Khare, Dan Connolly, and Tim Berners-Lee.
       Privacy and Demographics Overview.
       http://www.w3.org/pub/WWW/Demographics/.

   11. Linda Tauscher and Saul Greenberg.  Revisitation Patterns in
       World Wide Web Navigation.  Research Report 96/587/07, Department
       of Computer Science, University of Calgary, March, 1996.
       http://www.cpsc.ucalgary.ca/projects/grouplab/
       papers/96WebReuse/TechReport96.html.

   12. Wessels, D., and K. Claffy "Internet Cache Protocol (ICP),
       version 2", RFC 2186, September 1997.

13 Authors' Addresses



   Jeffrey C. Mogul
   Western Research Laboratory
   Digital Equipment Corporation
   250 University Avenue
   Palo Alto, California, 94305, U.S.A.

   EMail: mogul@wrl.dec.com
   Phone: 1 415 617 3304 (email preferred)


   Paul J. Leach
   Microsoft
   1 Microsoft Way
   Redmond, Washington, 98052, U.S.A.

   EMail: paulle@microsoft.com




















Mogul & Leach               Standards Track                    [Page 36]

RFC 2227            Hit-Metering and Usage-Limiting         October 1997


14 Full Copyright Statement



   Copyright (C) The Internet Society (1997).  All Rights Reserved.

   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implmentation may be prepared, copied, published
   andand distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph are
   included on all such copies and derivative works.  However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than
   English.

   The limited permissions granted above are perpetual and will not be
   revoked by the Internet Society or its successors or assigns.

   This document and the information contained herein is provided on an
   "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
   TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
   BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
   HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
























Mogul & Leach               Standards Track                    [Page 37]