RFC 1363






Network Working Group                                     C. Partridge
Request for Comments: 1363                                         BBN
                                                        September 1992


                     A Proposed Flow Specification

Status of this Memo



   This memo provides information for the Internet community.  It does
   not specify an Internet standard.  Distribution of this memo is
   unlimited.

Abstract



   A flow specification (or "flow spec") is a data structure used by
   internetwork hosts to request special services of the internetwork,
   often guarantees about how the internetwork will handle some of the
   hosts' traffic.  In the future, hosts are expected to have to request
   such services on behalf of distributed applications such as
   multimedia conferencing.

   The flow specification defined in this memo is intended for
   information and possible experimentation (i.e., experimental use by
   consenting routers and applications only).  This RFC is a product of
   the Internet Research Task Force (IRTF).

Introduction

   The Internet research community is currently studying the problems of
   supporting a new suite of distributed applications over
   internetworks.  These applications, which include multimedia
   conferencing, data fusion, visualization, and virtual reality, have
   the property that they require the distributed system (the collection
   of hosts that support the applications along with the internetwork to
   which they are attached) be able to provide guarantees about the
   quality of communication between applications.  For example, a video
   conference may require a certain minimum bandwidth to be sure that
   the video images are delivered in a timely way to all recipients.

   One way for the distributed system to provide guarantees is for hosts
   to negotiate with the internetwork for rights to use a certain part
   of the internetwork's resources.  (An alternative is to have the
   internetwork infer the hosts' needs from information embedded in the
   data traffic each host injects into the network.  Currently, it is
   not clear how to make this scheme work except for a rather limited
   set of traffic classes.)




Partridge                                                       [Page 1]

RFC 1363             A Proposed Flow Specification        September 1992


   There are a number of ways to effect a negotiation.  For example a
   negotiation can be done in-band or out-of-band.  It can also be done
   in advance of sending data (possibly days in advance), as the first
   part of a connection setup, or concurrently with sending (i.e., a
   host starts sending data and starts a negotiation to try to ensure
   that it will allowed to continue sending).  Insofar as is possible,
   this memo is agnostic with regard to the variety of negotiation that
   is to be done.

   The purpose of this memo is to define a data structure, called a flow
   specification or flow spec, that can be used as part of the
   negotiation to describe the type of service that the hosts need from
   the internetwork.  This memo defines the format of the fields of the
   data structure and their interpretation.  It also briefly describes
   what purpose the different fields fill, and discusses why this set of
   fields is thought to be both necessary and sufficient.

   It is important to note that the goal of this flow spec is to able to
   describe *any* flow requirement, both for guaranteed flows and for
   applications that simply want to give hints to the internetwork about
   their requirements.

Format of the Flow Spec

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |              Version          |    Maximum Transmission Unit  |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |      Token Bucket Rate        |        Token Bucket Size      |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  Maximum Transmission Rate    |     Minimum Delay Noticed     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |     Maximum Delay Variation   |        Loss Sensitivity       |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |     Burst Loss Sensitivity    |          Loss Interval        |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |    Quality of Guarantee       |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Discussion of the Flow Spec

   The flow spec indicates service requirements for a single direction.
   Multidirectional flows will need to request services in both
   directions (using two flow specs).

   To characterize a unidirectional flow, the flow spec needs to do four
   things.



Partridge                                                       [Page 2]

RFC 1363             A Proposed Flow Specification        September 1992


   First, it needs to characterize how the flow's traffic will be
   injected into the internetwork.  If the internetwork doesn't know
   what to expect (is it a gigabit-per-second flow or a three kilobit-
   per-second flow?) then it is difficult for the internetwork to make
   guarantees.  (Note the word "difficult" rather than "impossible."  It
   may be possible to statistically manage traffic or over-engineer the
   network so well that the network can accept almost all flows, without
   setup.  But this problem looks far harder than asking the sender to
   approximate its behavior so the network can plan.)  In this flow
   spec, injected traffic is characterized as having a sustainable rate
   (the token bucket rate) a peak rate (the maximum transmission rate),
   and an approximate burst size (the token bucket size).  A more
   precise definition of each of these fields is given below.  The
   characterization is based, in part, on the work done in [1].

   Second, the flow spec needs to characterize sensitivity to delay.
   Some applications are more sensitive than others.  At the same time,
   the internetwork will likely have a choice of routes with various
   delays available from the source to destination.  For example, both
   routes using satellites (which have very long delays) and routes
   using terrestrial lines (which will have shorter delays) may be
   available.  So the sending host needs to indicate the flow's
   sensitivity to delay.  However, this field is only advisory.  It only
   tells the network when to stop trying to reduce the delay - it does
   not specify a maximum acceptable delay.

   There are two problems with allowing applications to specify the
   maximum acceptable delay.

   First, observe that an application would probably be happy with a
   maximum delay of 100 ms between the US and Japan but very unhappy
   with a delay of 100 ms within the same city.  This observation
   suggests that the maximum delay is actually variable, and is a
   function of the delay that is considered achievable.  But the
   achievable delay is largely determined by the geographic distance
   between the two peers, and this sort of geographical information is
   usually not available from a network.  Worse yet, the advent of
   mobile hosts makes such information increasingly hard to provide.  So
   there is reason to believe that applications may have difficulty
   choosing a rational maximum delay.

   The second problem with maximum delays is that they are an attempt to
   quantify what performance is acceptable to users, and an application
   usually does not know what performance will be acceptable its user.
   For example, a common justification for specifying a maximum
   acceptable delay is that human users find it difficult to talk to
   each other over a link with more than about 100 ms of delay.
   Certainly such delays can make the conversation less pleasant, but it



Partridge                                                       [Page 3]

RFC 1363             A Proposed Flow Specification        September 1992


   is still possible to converse when delays are several seconds long,
   and given a choice between no connection and a long delay, many users
   will pick the delay.  (The phone call may involve an important matter
   that must be resolved.)

   As part of specifying a flow's delay sensitivity, the flow spec must
   also characterize how sensitive the flow is to the distortion of its
   data stream.

   Packets injected into a network according to some pattern will not
   normally come out of the network still conforming to the pattern.
   Instead, the pattern will have been distorted by queueing effects in
   the network.  Since there is reason to believe that it may make
   network design easier to continue to allow the networks slightly
   distort traffic patterns, it is expected that those applications
   which are sensitive to distortion will require their hosts to use
   some amount of buffering to reshape the flow back into its original
   form.  It seems reasonable to assume that buffer space is not
   infinite and that a receiving system will wish to limit the amount of
   buffering that a single flow can use.

   The amount of buffer space required for removing distortion at the
   receiving system is determined by the variation in end-to-end
   transmission delays for data sent over the flow.  If the transmission
   delay is a mean delay, D, plus or minus a variance, V, the receiving
   system needs buffer space equivalent to 2 * V * the transmission
   rate.  To see why this is so, consider two packets, A and B, sent T
   time units apart which must be delivered to the receiving application
   T time units apart.  In the worst case, A arrives after a delay of
   D-V time units (the minimum delay) and B arrives after a delay of D+V
   time units (the maximum delay).  The receiver cannot deliver B until
   it arrives, which is T + 2 * V time units after A.  To ensure that A
   is delivered T time units before B, A must be buffered for 2 * V time
   units.  The delay variance field is the value of 2 * V, and allows
   the receiver to indicate how much buffering it is willing to provide.

   A third function of the flow spec is to signal sensitivity to loss of
   data.  Some applications are more sensitive to the loss of their data
   than other applications.  Some real-time applications are both
   sensitive to loss and unable to wait for retransmissions of data.
   For these particularly sensitive applications, hosts may implement
   forward error correction on a flow to try to absolutely minimize
   loss.  The loss fields allow hosts to request loss properties
   appropriate for the application's requirements.

   Finally, it is expected that the internetwork may be able to provide
   a range of service guarantees.  At the best, the internetwork may be
   asked to guarantee (with tight probability bounds) the quality of



Partridge                                                       [Page 4]

RFC 1363             A Proposed Flow Specification        September 1992


   service it will provide.  Or the internetwork may simply be asked to
   ensure that packets sent over the flow take a terrestrial path.  The
   quality of guarantee field indicates what type of service guarantee
   the application desires.

Definition of Individual Fields

General Format of Fields

   With a few exceptions, fields of the flow spec are expressed using a
   common 16-bit format.  This format has two forms.  The first form is
   shown below.

               0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
              +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
              |0|  Exponent   |     Value     |
              +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

   In this format, the first bit is 0, followed by 7 bits of an exponent
   (E), and an 8-bit value (V).  This format encodes a number, of the
   form V * (2**E).  This representation was chosen to allow easy
   representation of a wide range of values, while avoiding over-precise
   representations.

   In some case, systems will not wish to request a precise value but
   rather simply indicate some sensitivity.  For example, a virtual
   terminal application like Telnet will likely want to indicate that it
   is sensitive to delay, but it may not be worth expressing particular
   delay values for the network to try to achieve.  For these cases,
   instead of a number, the field in the flow spec will take the
   following form:

               0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
              +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
              |1|   Well-defined Constant     |
              +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

   The first bit of the field is one, and is followed by a 15-bit
   constant.  The values of the constants for given fields are defined
   below.  Any additional values can be requested from the Internet
   Assigned Numbers Authority (IANA).

   Version Field

      This field is a 16-bit integer in Internet byte order.  It is the
      version number of the flow specification.  The version number of
      the flow specification defined in this document is 1.  The IANA is
      responsible for assigning future version numbers for any proposed



Partridge                                                       [Page 5]

RFC 1363             A Proposed Flow Specification        September 1992


      revisions of this flow specification.

      This field does not use the general field format.

   Maximum Transmission Unit (MTU)

      A 16-bit integer in Internet byte order which is the maximum
      number of bytes in the largest possible packet to be transmitted
      over this flow.

      This field does not use the general field format.

      The field serves two purposes.

      It is a convenient unit for expressing loss properties.  Using the
      default MTU of the internetwork is inappropriate since the
      internetwork have very large MTU, such the 64Kbytes of IP, but
      applications and hosts may be sensitive to losses of far less than
      an MTU's amount of data -- for example, a voice application would
      be sensitive to a loss of several consecutive small packets.

      The MTU also bounds the amount of time that a flow can transmit,
      uninterrupted, on a shared media.

      Similarly, the loss rates of links that suffer bit errors will
      vary dramatically based on the MTU size.

   Token Bucket Rate

      The token bucket rate is one of three fields used to define how
      traffic will be injected into the internetwork by the sending
      application.  (The other two fields are the token bucket size and
      the maximum transmission rate.)

      The token rate is the rate at which tokens (credits) are placed
      into an imaginary token bucket.  For each flow, a separate bucket
      is maintained.  To send a packet over the flow, a host must remove
      a number of credits equal to the size of the packet from the token
      bucket.  If there are not enough credits, the host must wait until
      enough credits accumulate in the bucket.

      Note that the fact that the rate is expressed in terms of a token
      bucket rate does not mean that hosts must implement token buckets.
      Any traffic management scheme that yields equivalent behavior is
      permitted.

      The field is in the general field format and counts the number of
      byte credits (i.e., right to send a byte) per second which are



Partridge                                                       [Page 6]

RFC 1363             A Proposed Flow Specification        September 1992


      deposited into the token bucket.  The value must be a number (not
      a well-known constant).

      The value zero is slightly special.  It is used to indicate that
      the application is not making a request for bandwidth guarantees.
      If this field is zero, then the Token Bucket Size must also be
      zero, and the type of guarantee requested may be no higher than
      predicted service.

   Token Bucket Size

      The token bucket size controls the maximum amount of data that the
      flow can send at the peak rate.  More formally, if the token
      bucket size is B, and the token bucket rate is R, over any
      arbitrarily chosen interval T in the life of the flow, the amount
      of data that the flow sends cannot have exceeded B + (R * T)
      bytes.

      The token bucket is filled at the token bucket rate.  The bucket
      size limits how many credits the flow may store.  When the bucket
      is full, new credits are discarded.

      The field is in the general field format and indicates the size of
      the bucket in bytes.  The value must be a number.

      Note that the bucket size must be greater than or equal to the MTU
      size.

      Zero is a legal value for the field and indicates that no credits
      are saved.

   Maximum Transmission Rate

      The maximum transmission rate limits how fast packets may be sent
      back to back from the host.  Consider that if the token bucket is
      full, it is possible for the flow to send a series of back-to-back
      packets equal to the size of the token bucket.  If the token
      bucket size is large, this back-to-back run may be long enough to
      significantly inhibit multiplexing.

      To limit this effect, the maximum transmission rate bounds how
      fast successive packets may be placed on the network.

      One can think of the maximum transmission rate control as being a
      form of a leaky bucket.  When a packet is sent, a number of
      credits equal to the size of the packet is placed into an empty
      bucket, which drains credits at the maximum transmission rate.  No
      more packets may be sent until the bucket has emptied again.



Partridge                                                       [Page 7]

RFC 1363             A Proposed Flow Specification        September 1992


      The maximum transmission rate is the rate at which the bucket is
      emptied.  The field is in the general field format and indicates
      the size of the bucket in bytes.  The value must be a number and
      must be greater than or equal to the token bucket rate.

      Note that the MTU size can be used in conjunction with the maximum
      transmission rate to bound how long an individual packet blocks
      other transmissions.  The MTU specifies the maximum time an
      individual packet may take.  The Maximum Transmission Rate, limits
      the frequency with which packets may be placed on the network.

   Minimum Delay Noticed

      The minimum delay noticed field tells the internetwork that the
      host and application are effectively insensitive to improvements
      in end-to-end delay below this value.  The network is encouraged
      to drive the delay down to this value but need not try to improve
      the delay further.

      The field is in the general field format.

      If expressed as a number it is the number of microseconds of delay
      below which the host and application do not care about
      improvements.  Human users only care about delays in the
      millisecond range but some applications will be computer to
      computer and computers now have clock times measured in a handful
      of nanoseconds.  For such computers, microseconds are an
      appreciable time.  For this reason, this field measures in
      microseconds, even though that may seem small.

      If expressed as a well-known constant (first bit set), two field
      values are accepted:

         0 - the application is not sensitive to delay



         1 - the application is moderately delay sensitive
             e.g., avoid satellite links where possible).



   Maximum Delay Variation

      If a receiving application requires data to be delivered in the
      same pattern that the data was transmitted, it may be necessary
      for the receiving host to briefly buffer data as it is received so
      that the receiver can restore the old transmission pattern.  (An
      easy example of this is a case where an application wishes to send
      and transmit data such as voice samples, which are generated and
      played at regular intervals.  The regular intervals may be
      distorted by queueing effects in the network and the receiver may



Partridge                                                       [Page 8]

RFC 1363             A Proposed Flow Specification        September 1992


      have to restore the regular spacing.)

      The amount of buffer space that the receiving host is willing to
      provide determines the amount of variation in delay permitted for
      individual packets within a given flow.  The maximum delay
      variation field makes it possible to tell the network how much
      variation is permitted.  (Implementors should note that the
      restrictions on the maximum transmission rate may cause data
      traffic patterns to be distorted before they are placed on the
      network, and that this distortion must be accounted for in
      determining the receiver buffer size.)

      The field is in the general field format and must be a number.  It
      is the difference, in microseconds, between the maximum and
      minimum possible delay that a packet will experience.  (There is
      some question about whether microsecond units are too large.  At a
      terabit per second, one microsecond is a megabit.  Presumably if a
      host is willing to receive data at terabit speeds it is willing to
      provide megabits of buffer space.)

      The value of 0, meaning the receiving host will not buffer out
      delays, is acceptable but the receiving host must still have
      enough buffer space to receive a maximum transmission unit sized
      packet from the sending host.  Note that it is expected that a
      value of 0 will make it unlikely that a flow can be established.

   Loss Sensitivity

      This field indicates how sensitive the flow's traffic is to
      losses.  Loss sensitivity can be expressed in one of two ways:
      either as a number of losses of MTU-sized packets in an interval,
      or simply as a value indicating a level of sensitivity.

      The field is in the general field format.

      If the value is a number, then the value is the number of MTU-
      sized packets that may be lost out of the number of MTU-sized
      packets listed in the Loss Interval field.

      If the value is a well-known constant, then one of two values is
      permitted:

         0 - the flow is insensitive to loss

         1 - the flow is sensitive to loss (where possible
             choose the path with the lowest loss rate).





Partridge                                                       [Page 9]

RFC 1363             A Proposed Flow Specification        September 1992


   Burst Loss Sensitivity

      This field states how sensitive the flow is to losses of
      consecutive packets.  The field enumerates the maximum number of
      consecutive MTU-sized packets that may be lost.

      The field is in the general field format.

      If the value is a number, then the value is the number of
      consecutive MTU-sized packets that may be lost.

      If the value is a well-known constant, then the value 0 indicates
      that the flow is insensitive to burst loss.

      Note that it is permissible to set the loss sensitivity field to
      simply indicate sensitivity to loss, and set a numerical limit on
      the number of consecutive packets that can be lost.

   Loss Interval

      This field determines the period over which the maximum number of
      losses per interval are measured.  In other words, given any
      arbitrarily chosen interval of this length, the number of losses
      may not exceed the number in the Loss Sensitivity field.

      The field is in the general field format.

      If the Loss Sensitivity field is a number, then this field must
      also be a number and must indicate the number of MTU-sized packets
      which constitutes a loss interval.

      If the Loss Sensitivity field is not a number (i.e., is a well-
      known constant) then this field must use the well-known constant
      of 0 (i.e., first bit set, all other bits 0) indicating that no
      loss interval is defined.

   Quality of Guarantee

      It is expected that the internetwork will likely have to offer
      more than one type of guarantee.

      There are two unrelated issues related to guarantees.

      First, it may not be possible for the internetwork to make a firm
      guarantee.  Consider a path through an internetwork in which the
      last hop is an Ethernet.  Experience has shown (e.g., some of the
      IETF conferencing experiments) that an Ethernet can often give
      acceptable performance, but clearly the internetwork cannot



Partridge                                                      [Page 10]

RFC 1363             A Proposed Flow Specification        September 1992


      guarantee that the Ethernet will not saturate at some time during
      a flow's lifetime.  Thus it must be possible to distinguish
      between flows which cannot tolerate the small possibility of a
      failure (and thus must guaranteed at every hop in the path) and
      those that can tolerate islands of uncertainty.

      Second, there is some preliminary work (see [2]) that suggests
      that some applications will be able to adapt to modest variations
      in internetwork performance and that network designers can exploit
      this flexibility to allow better network utilization.  In this
      model, the internetwork would be allowed to deviate slightly from
      the promised flow parameters during periods of load.  This class
      of service is called predicted service (to distinguish it from
      guaranteed service).

      The difference between predicted service and service which cannot
      be perfectly guaranteed (e.g., the Ethernet example mentioned
      above) is that the imperfect guarantee makes no statistical
      promises about how it might mis-behave.  In the worst case, the
      imperfect guarantee will not work at all, whereas predicted
      service will give slightly degraded service.  Note too that
      predicted service assumes that the routers and links in a path all
      cooperate (to some degree) whereas an imperfect guarantee states
      that some routers or links will not cooperate.

      The field is a 16-bit field in Internet byte order.  There are six
      legal values:

         0 - no guarantee is required (the host is simply expressing
             desired performance for the flow)

         100 (hex) - an imperfect guarantee is requested.



         200 (hex) - predicted service is requested and if unavailable,
                     then no flow should be established.



         201 (hex) - predicted service is requested but an imperfect
                     guarantee is acceptable.



         300 (hex) - guaranteed service is requested and if a firm
                     guarantee cannot be given, then no flow should be
                     established.



         301 (hex) - guaranteed service is request and but an imperfect
                     guarantee is acceptable.



      It is expected that asking for predicted service or permitting an
      imperfect guarantee will substantially increase the chance that a



Partridge                                                      [Page 11]

RFC 1363             A Proposed Flow Specification        September 1992


      flow request will be accepted.

Possible Limitations in the Proposed Flow Spec

   There are at least three places where the flow spec is arguably
   imperfect, based on what we currently know about flow reservation.
   In addition, since this is a first attempt at a flow spec, readers
   should expect modifications as we learn more.

   First, the loss model is not perfect.  Simply stating that an
   application is sensitive to loss and to burst loss is a rather crude
   indication of sensitivity.  However, explicitly enumerating loss
   requirements within a cycle is also an imperfect mechanism.  The key
   problem with the explicit values is that not all packets sent over a
   flow will be a full MTU in size.  Expressed another way, the current
   flow spec expects that an MTU-sized packet will be the unit of error
   recovery.  If flows send packets in a range of sizes, then the loss
   bounds may not be very useful.  However, the thought of allowing a
   flow to request a set of loss models (one per packet size) is
   sufficiently painful that I've limited the flow to one loss profile.
   Further study of loss models is clearly needed.

   Second, the minimum delay sensitivity field limits a flow to stating
   that there is one point on a performance sensitivity curve below
   which the flow is no longer interested in improved performance.  It
   may be that a single point is insufficient to fully express a flow's
   sensitivity.  For example, consider a flow for supporting part of a
   two-way voice conversation.  Human users will notice improvements in
   delay down to a few 10s of milliseconds.  However, the key point of
   sensitivity is the delay at which normal conversation begins to
   become awkward (about 100 milliseconds).  By allowing only one
   sensitivity point, the flow spec forces the flow designer to either
   ask for the best possible delay (e.g, a few 10's of ms) to try to get
   maximum performance from the network, or state a sensitivity of about
   95 ms, and accept the possibility that the internetwork will not try
   to improve delay below that value, even if it could (and even though
   the user would notice the improvement).  My expectation is that a
   simple point is likely to be easier to deal with than attempting to
   enumerate two (or three or four) points in the sensitivity curve.

   Third, the models for service guarantees is still evolving and it is
   by no means clear that the service choices provided are the correct
   set.








Partridge                                                      [Page 12]

RFC 1363             A Proposed Flow Specification        September 1992


How an Internetwork is Expected to Handle a Flow Spec

   There are at least two parts to the issue of how an internetwork is
   expected to handle a flow spec.  The first part deals with how the
   flow spec is interpreted so that the internetwork can find a route
   which will allow the internetwork to match the flow's requirements.
   The second part deals with how the network replies to the host's
   request.

   The precise mechanism for setting up a flow, given a flow spec, is a
   large topic and beyond the scope of this memo.  The purpose of the
   next few paragraphs is simply to sketch an argument that this flow
   spec is sufficient to the requirements of the setup mechanisms known
   to the author.

   The key problem in setting up a flow is determining if there exist
   one or more routes from the source to the destination(s) which might
   be able to support the quality of service requested.  Once one has a
   route (or set of candidate routes) one can take whatever actions may
   be appropriate to confirm that the route is actually viable and to
   cause the flow's data to follow that route.

   There are a number of ways to find a route.  One might try to build a
   route on the fly by establishing the flow hop-by-hop (as ST-II does)
   or one might consult a route server which provides a set of candidate
   source routes derived from a routing database.  However, whatever
   system is used, some basic information about the flow needs to be
   provided to the routing system.  This information is:

      * How much bandwidth the flow may require.  There's no point
        in routing a flow that expects to send at over 10 megabits per
        second via a T1 (1.5 megabit per second) link.

      * How delay sensitive the application is.  One does not wish
        to route a delay-sensitive application over a satellite link,
        unless the satellite link is the only possible route from here
        to there.

      * How much error can be tolerated.  Can we send this flow over
        our microwave channel on a rainy day or is a more reliable link
        required?

      * How firm the guarantees need to be.  Can we put an Ethernet
        in as one of the hops?

      * How much delay variation is tolerated.  Again, can an Ethernet
        be included in the path?  Does the routing system need to worry
        if the addition of this flow will cause a few routers to run



Partridge                                                      [Page 13]

RFC 1363             A Proposed Flow Specification        September 1992


        at close to capacity?  (A side note: we assume that the routers
        are running with priority queueing systems, so running the router
        close to capacity doesn't mean that all flows get long and
        variable delays.  Rather, running close to capacity means that
        high priority flows will be unaffected, and low priority flows
        will get hit with a lot of delay and variation.)

   The flow spec provides all of this information.  So it seems
   plausible to assume it provides enough information to make routing
   decisions at setup time.

   The flow spec was designed with the expectation that the network
   would give a yes or no reply to a request for a guaranteed flow.

   Some researchers have suggested that the negotiation to set up a flow
   might be an extended negotiation, in which the requesting host
   initially requests the best possible flow it could desire and then
   haggles with the network until they agree on a flow with properties
   that the network can actually provide and the application still finds
   useful.  This notion bothers me for at least two reasons.  First, it
   means setting up a flow is a potentially long process.  Second, the
   general problem of finding all possible routes with a given set of
   properties is a version of the traveling salesman problem, and I
   don't want to embed traveling salesman algorithms into a network's
   routing system.

   The model used in designing this flow spec was that a system would
   ask for the minimum level of service that was deemed acceptable and
   the network would try to find a route that met that level of service.
   If the network is unable to achieve the desired level of service, it
   refuses the flow, otherwise it accepts the flow.

The Flow Spec as a Return Value

   This memo does not specify the data structures that the network uses
   to accept or reject a flow.  However, the flow spec has been designed
   so that it can be used to return the type of service being
   guaranteed.

   If the request is being accepted, the minimum delay field could be
   set to the guaranteed or predicted delay, and the quality of
   guarantee field could be set to no guarantee (0), imperfect guarantee
   (100 hex), predicted service (200 hex), or guaranteed service (300
   hex).

   If the request is being rejected, the flow spec could be modified to
   indicate what type of flow the network believes it could accept e.g.,
   the traffic shape or delay characteristics could be adjusted or the



Partridge                                                      [Page 14]

RFC 1363             A Proposed Flow Specification        September 1992


   type of guarantee lowered).  Note that this returned flow spec would
   likely be a hint, not a promised offer of service.

Why Type of Service is not Good Enough

   The flow spec proposed in this memo takes the form of a set of
   parameters describing the properties and requirements of the flow.
   An alternative approach which is sometimes mentioned (and which is
   currently incorporated into IP) is to use a Type of Service (TOS)
   value.

   The TOS value is an integer (or bit pattern) whose values have been
   predefined to represent requested quality of services.  Thus, a TOS
   of 47 might request service for a flow using up to 1 gigabit per
   second of bandwidth with a minimum delay sensitivity of 100
   milliseconds.

   TOS schemes work well if the different quality of services that may
   be requested are both enumerable and reasonably small.
   Unfortunately, these conditions do not appear to apply to future
   internetworks.  The range of possible bandwidth requests alone is
   huge.  Combine this range with several gradations of delay
   requirements, and widely different sensitivities to errors and the
   set of TOS values required becomes extremely large.  (At least one
   person has suggested to the author that perhaps a TOS field combined
   with a bandwidth parameter might be appropriate.  In other words, a
   two parameter model.  That's a tempting idea but my gut feeling is
   that it is not quite sufficient so I'm proposing a more complete
   parametric model.)

   Another reason to prefer parametric service is optimization issues.
   A key issue in flow setup is trying to design the the routing system
   to optimize its management of flows.  One can optimize on a number of
   criteria.  A good example of an optimization problem is the following
   question (expressed by Isidro Castineyra of BBN):

     "Given a request to establish a flow, how can the internetwork
     accept that request in such a way as to maximize the chance that
     the internetwork will also be able to accept the next flow
     request?"

   The optimization goal here is call-completion - maximizing the chance
   that requests to establish flows will succeed.  One might
   alternatively try to maximize revenue (if one is charging for flows).

   The internetwork is presumably in a better position to do
   optimizations if it has more information about the flow's expected
   behavior.  For example, if a TOS system says only that a flow is



Partridge                                                      [Page 15]

RFC 1363             A Proposed Flow Specification        September 1992


   delay sensitive, the routing system must seek out the most direct
   route for the flow.  But if the routing system is told that the flow
   is sensitive only to delays over 100 milliseconds, there may be a
   number of routes other than the most direct route which can satisfy
   this delay, thus leaving the most direct route available for a later
   flow which needs a far lower delay.

   In fairness, it should be noted that a danger of a parametric model
   is that it is very easy to have too many parameters.  The yearn to
   optimize can be overdone.  The goal of this flow spec is to enumerate
   just enough parameters that it appears that essential needs can be
   expressed, and the internetwork has some information it can use to
   try to manage the flows.  Features that would simply be nice or
   useful to have (but not essential) are left out to keep the parameter
   space small.

An Implication of the Flow Spec

   It is important to observe that the there are fields in the flow spec
   that are based on information from the sender (such as rate
   information) and fields in the flow spec that are based on
   information from the receiver (such as delay variation).  There are
   also fields that may sender and receiver to negotiate in advance.
   For example, the acceptable loss rate may depend on whether the
   sender and receiver both support the same type of forward error
   correction.  The delay sensitivity for a voice connection may depend,
   in part, on whether both sender and receiver support echo cancelling.

   The implication is that the internetwork must permit the sender and
   receiver to communicate in advance of setting up a flow, because a
   flow spec can only be defined once both sender and receiver have had
   their say.  In other words, a reserved flow should not be the only
   form of communication.   There must be some mechanism to perform a
   short exchange of messages in preparation for setting up a flow.

   (Another aside: it has been suggested that perhaps the solution to
   this problem is to have the sender establish a flow with an
   incomplete flow spec, and when the receiver gets the flow spec, have
   the receiver send the completed flow spec back along the flow, so the
   internetwork can "revise" the flow spec according to the receiver's
   desires.  I have two problems with this approach.  First, it is
   entirely possible that the receiver's information may lead the
   internetwork to conclude that the flow established by the sender is
   no good.  For example, the receiver may indicate it has a smaller
   tolerance for delay variation than expected and force the flow to be
   rerouted over a completely different path.  Second, if we try to
   avoid having the receiver's information cause the flow to fail, then
   we have to over-allocate the flow's during the preliminary setup.



Partridge                                                      [Page 16]

RFC 1363             A Proposed Flow Specification        September 1992


   But over allocating the resources requested may lead us to choose
   better quality paths than we need for this flow.  In other words, our
   attempts to optimize use of the network will fail.)

Advance Reservations and Flow Duration

   The primary purpose of a flow specification is to provide information
   to the internetwork so the internetwork can properly manage the
   proposed flow's traffic in the context of other traffic in the
   internetwork.  One question is whether the flow should give the
   network information about when the flow is expected to start and how
   long the flow is expected to last.

   Announcing when a flow will start is generally of interest for
   advance reservations.  (If the flow is not be reserved substantially
   in advance, the presentation of the flow spec to the internetwork can
   be taken as an implicit request for a flow, now.)  It is my view that
   advance reservation is a distinct problem from the describing the
   properties of a flow.  Advanced reservations will require some
   mechanism to maintain information in the network about flows which
   are not currently active but are expected to be activated at some
   time in the future.  I anticipate this will require some sort of
   distributed database to ensure that information about advanced
   reservations is not accidentally lost if parts of the internetwork
   crash.  In other words, advance reservations will require
   considerable additional supporting baggage that it would probably be
   better to keep out of the average flow spec.

   Deciding whether a flow spec should contain information about how
   long the flow is expected to run is a harder decision to make.
   Clearly if we anticipate that the internetwork will support advance
   reservations, it will be necessary for elements of the internetwork
   to predict their traffic load, so they can ensure that advance
   reservations are not compromised by new flow requests.  However,
   there is a school of thought that believes that estimating future
   load from current behavior of existing flows is more accurate than
   anything the flows may have declared in their flow specs.  For this
   reason, I've left a duration field out of the flow spec.

Examples

   To illustrate how the flow spec values might be used, this section
   presents three example flow specs.

   Telnet

      For the first example, consider using the flow spec to request
      service for an existing application: Telnet.  Telnet is a virtual



Partridge                                                      [Page 17]

RFC 1363             A Proposed Flow Specification        September 1992


      terminal protocol, and one can think of it as stringing a virtual
      wire across the network between the user's terminal and a remote
      host.

      Telnet has proved a very successful application without a need to
      reserve bandwidth: the amount of data sent over any Telnet
      connection tends to be quite small.  However, Telnet users are
      often quite sensitive to delay, because delay can affect the time
      it takes to echo characters.  This suggests that a Telnet
      connection might benefit from asking the internetwork to avoid
      long delay paths.  It could so so using the following flow spec
      (for both directions):

      Version=1
      MTU=80 [40 bytes of overhead + 40 bytes user data]
      Token Bucket Rate=0/0/0 [don't want a guarantee]
      Token Bucket Size=0/0/0
      Maximum Transmission Rate=0/0/0
      Maximum Delay Noticed=1/1 [constant = delay sensitive]
      Maximum Delay Variation=0/0/0 [not a concern]
      Loss Sensitivity=1/0 [don't worry about loss]
      Burst Loss Sensitivity=1/0
      Loss Interval=1/0
      Quality of Guarantee=1/0 [just asking]

      It is worth noting that Telnet's flow spec is likely to be the
      same for all instantiations of a Telnet connection.  As a result,
      there may be some optimizations possible (such as just tagging
      Telnet packets as being subject to the well-known Telnet flow
      spec).

   A Voice Flow

      Now consider transmitting voice over the Internet.  Currently,
      good quality voice can be delivered at rates of 32Kbit/s or
      16Kbit/s.  Assuming the rate is 32Kbit/s and voice samples are 16
      bit samples packaged into UDP datagrams (for a data rate of about
      60 Kbyte/s), a flow spec might be:

      Version=1
      MTU=30 [2 byte sample in UDP datagram]
      Token Bucket Rate=0/10/59 [60.4 Kbytes/s]
      Token Bucket Size=0/0/30 [save enough to send immediately
                                after pauses]
      Maximum Transmission Rate=0/10/59 [peak same as mean]
      Maximum Delay Noticed=0/10/100 [100 ms]
      Maximum Delay Variation=0/10/10 [keep variation low]
      Loss Sensitivity=1/1 [loss sensitive]



Partridge                                                      [Page 18]

RFC 1363             A Proposed Flow Specification        September 1992


      Burst Loss Sensitivity=0/0/5 [keep bursts small]
      Loss Interval=1/0
      Quality of Guarantee=1/201 [predicted service and I'll accept
                                  worse]

   A Variable Bit-Rate Video Flow

      Variable bit-rate video transmissions vary the rate at which they
      send data according to the amount of the video image that has
      changed between frames.  In this example, we consider a one-way
      broadcast of a picture.  If we assume 30 frames a second and that
      a full frame is about 1 megabit of data, and that on average about
      10% of the frame changes, but in the worst case the entire frame
      changes, the flow spec might be:

      Version=1
      MTU=4096 [big so we can put lots of bits in each packet]
      Token Bucket Rate=0/20/1 [8 Mbits/s]
      Token Bucket Size=0/17/2 [2 Mbits/s]
      Maximum Transmission Rate=0/20/30 [30 Mbits/s]
      Maximum Delay Noticed=1/1 [somewhat delay sensitive]
      Maximum Delay Variation=0/10/1 [no more than one second of
                                      buffering]
      Loss Sensitivity=0/0/1 [worst case, one loss per frame]
      Burst Loss Sensitivity=0/0/1 [no burst errors please]
      Loss Interval=0/0/33 [one frame in MTU sized packets]
      Quality of Guarantee=1/300 [guaranteed service only]

      The token bucket is sized to be two frames of data, and the bucket
      rate will fill the bucket every 250 ms.  The expectation is that
      full scene changes will be rare and that a fast rate with a large
      bucket size should accommodate even a series of scene changes.

   Disclaimer

      In all cases, these examples are simply to sketch the use of the
      flow spec.  The author makes no claims that the actual values used
      are the correct ones for a particular application.

Security Considerations

   Security considerations definitely exist.  For example, one might
   assume that users are charged for guaranteed flows.  In that case,
   some mechanism must exist to ensure that a flow request (including
   flow spec) is authenticated.  However I believe that such issues have
   to be dealt with as part of designing a negotiation protocol, and are
   not part of designing the flow spec data structure.




Partridge                                                      [Page 19]

RFC 1363             A Proposed Flow Specification        September 1992


Acknowledgements



   I'd like to acknowledge the tremendous assistance of Steve Deering,
   Scott Shenker and Lixia Zhang of XEROX PARC in writing this RFC.
   Much of this flow spec was sketched out in two long meetings with
   them at PARC.  Others who have offered notable advice and comments
   include Isidro Castineyra, Deborah Estrin, and members of the End-
   to-End Research Group chaired by Bob Braden.  All ideas that prove
   misbegotten are the sole responsibility of the author.  This work was
   funded under DARPA Contract No. MDA903-91-D-0019.  The views
   expressed in this document are not necessarily those of the Defense
   Advanced Research Projects Agency.

References

   1. Parekh, A., "A Generalized Processor Sharing Approach
      to Flow Control in Integrated Services Networks",
      MIT Laboratory for Information and Decision Systems,
      Report No. LIDS-TH-2089.

   2. Clark, D., Shenker, S., and L. Zhang, "Supporting Real-Time
      Applications in an Integrated Services Packet Network:
      Architecture and Mechanism", Proceedings of ACM SIGCOMM '92,
      August 1992.

Author's Address



   Craig Partridge
   BBN
   824 Kipling St
   Palo Alto, CA  94301

   Phone: 415-325-4541

   EMail: craig@aland.bbn.com
















Partridge                                                      [Page 20]