Network Working Group K. Smith Request For Comments: 1934 Ascend Communications Category: Informational April 1996
Ascend's Multilink Protocol Plus (MP+)
Status of This Memo
This memo provides information for the Internet community. This memo does not specify an Internet standard of any kind. Distribution of this memo is unlimited.
Abstract
This document proposes an extension to the PPP Multilink Protocol (MP) [1]. Multilink Protocol Plus (MP+) is a new control protocol for managing multiple data links that are bundled by MP.
The PPP Multilink Protocol (MP), is a set of features that provide inverse multiplexing at the packet/fragment level by bundling multiple independent links between a fixed pair of systems, providing a virtual link with greater bandwidth than any of the constituent members.
Once multiple channels have been established MP is responsible for managing channel use to insure in-sequence delivery of user packets.
MP+ is an extension to MP that adds an inband control channel to provide a new level of session management and control.
MP+ also allows remote device management of (unconfigured) systems. This feature allows a network operations center to dial into an unconfigured system and remotely manage it, before ethernet interface, IP address, and other LCP and system configuration information is entered. (This does require local configuration of the WAN interfaces to the extent required to answer an incoming call).
* Ability to negotiate to add and subtract channels when bandwidth needs change.
* Phone number management so calling stations need not know every possible number; answering stations can manage their own resources.
* A simple remote management interface.
Smith Informational [Page 2]
RFC 1934 Multilink Protocol Plus April 1996
To perform the above functions MP+ is split into a call management layer and a reliable delivery layer. The call management layer is the source and sink of MP+ control messages. The reliable delivery layer adds a simple acknowledge and retry mechanism.
MP+ only takes network bandwidth when in the process of performing a user request, e.g. adding and subtracting bandwidth.
NOTE: Neither MP, or MP+ define the process that makes the bandwidth requirement determination. That is outside the scope of either of these protocols and will likely be implementation dependent.
PPP In order to establish communications over a point-to-point link, each end of the PPP [2] link must first send LCP packets to configure the data link during link establishment phase. After the link has been established, PPP provides for an authentication phase.
MP The goal of multilink operation is to bundle multiple independent links between a fixed pair of systems, providing a virtual link with greater bandwidth than any of the constituent members.
MP+ MP+ is also negotiated during initial LCP option negotiation. A system indicates to its peer that it is willing to do MP+ by sending the MP+ option as part of the initial LCP option negotiation. The MP+ option MUST NOT be negotiated unless MP is also negotiated. When used, MP+ adds a virtual unit-to-unit control channel.
A peer may elect to:
Smith Informational [Page 3]
RFC 1934 Multilink Protocol Plus April 1996
Acknowledge both the MP and MP+ options, indicating that both MP and MP+ will be used.
Acknowledge the MP option and reject the MP+ option. Operation will fall back to MP.
Reject both options. Standard PPP will be used for this connection.
Standard PPP In standard PPP the LCP negotiation phase is followed by an optional authentication phase, and then one or more NCPs are initiated.
PPP with MP The LCP negotiation phase and authentication phase are identical to standard PPP. The ability to initiate an MP aggregate data link is indicated by sending an MP option - as described in [1].
PPP with MP and MP+ When MP+ is negotiated at LCP startup, the same procedures are followed as when MP is negotiated alone. The MP+ LCP option is negotiated to indicate the ability to use the MP+ feature.The first connection between endpoints causes the MP+ process to be started for the connection.
Figure 2: Multilink Plus Frame Format (short sequence number format)
MP+ frames use a similar structure to MP fragments.
The MP+ assigned PID is designated 00 73.
MP+ control uses the following two rules:
- MP+ control frames have their own sequence number space, controlled by MP+.
- MP+ control frames MUST NOT be fragmented.
NOTE: Implementations of this protocol prior to the date of submission of this specification to the IETF use the same PID as MP, but sets the LSB of the reserved bits in the MP header to 1 - this is how the MP+ packets are discriminated from MP fragments. So the header of the MP+ packet looks like:
00 3d c1 ......
As compared to an MP packet that looks like:
00 3d c0 ...... or 00 3d 80 ...... or 00 3d 40 ......
The error control layer that runs over the virtual inband channel is as simple as it can get, while handling the possibility of errors on the line.
An assumption is made that errors are infrequent, and that at the same time messages are rarely, if ever, dropped on the floor. The implication of this is that "timing out" on retransmission of messages does no harm. If a message cannot get through, then it simply is retried some number of times. After giving up, the only recourse is to notify the call management layer (of MP) that the session has died.
+---------------+---------------+ PPP Header: | Address 0xff | Control 0x03 | +---------------+---------------+ | PID(H) 0x00 | PID(L) 0x73 | +-+-+-+-+-------+---------------+ MP+ Header: |1|1|0|0|0|0|0|1| seq # high | +-+-+-+-+-------+---------------+ | sequence number low bits | +---------------+---------------+ EC Header: | Error Control Message Type | | 32 bits reserved | +---------------+---------------+ MP+ Data: | MP+ Message | May not be | | present.
Figure 3: MP+ control message format (shown long sequence number format)
Error Control Message Type:
1 DATA_MSG: This message contains MP+ data transferred between the peers.
2 ACK_MSG: An acknowledgement of a previous data message.
When set to DATA_MSG, the remainder of the frame contains an MP+ Control message.
When set to ACK_MSG, the remainder of the frame consists only of the PPP Frame Check Sum (FCS).
This layer is controlled by a simple state machine. There are three states:
Stopped There is no connection between peers.
Idle There is a connection between peers; no unacknowledged messages pending.
Pending There is a connection between peers; awaiting an acknowledgement to the last message sent.
Messages from the call management layer are queued for transmission whenever the link is in the pending state. For simplicity, only one outstanding message may be in the link at any given time. The entire procedure is defined in table 1.
Event State ______________________________________________________________________ Stopped Idle Pending ====================================================================== Start 1,Idle -,* -,* ______________________________________________________________________ Received ACK_MSG ** 2,Start 5,Idle|Pending current tx sequence number ______________________________________________________________________ Received ACK_MSG ** -,* -,* last tx sequence number ______________________________________________________________________ Received ACK_MSG ** 2,Start 2,Start other tx sequence number ______________________________________________________________________ Received DATA_MSG ** 6,* 6,* current rx sequence number ______________________________________________________________________ Received DATA_MSG ** 7,* 7,* previous rx sequence number ______________________________________________________________________ Received DATA_MSG ** 2,Start 2,Start other rx sequence number ______________________________________________________________________ Receive Invalid Frame ** 2,Start 2,Start ______________________________________________________________________ Retransmit Timer Expire ** ** 4,Start|* ______________________________________________________________________
** Invalid or meaningless event for state, ignored.
Notes:
[1] Data from the call management layer will always be copied before being queued for transmission. The call management layer is responsible for its own buffers.
[2] MP always copies data for transmission and returns immediately. Any buffers allocated to build control messages MUST be released immediately upon return from MP transmission requests.
Actions:
1 Reset rx sequence number Reset tx sequence number Reset tx retransmit count Stop retransmit timer
2 Report error to user Stop retransmit timer Stop frame transmit timer Free buffers
3 Save call management message in pending transmit queue Build DATA_MSG from first message in pending transmit queue using current tx sequence number. Send message to MP for transmission. Reset tx retransmit count
Smith Informational [Page 8]
RFC 1934 Multilink Protocol Plus April 1996
4 Increment tx retransmit count If tx retransmit count >= RETRANSMIT_COUNT Action 2 (followed by state change to the Start state) else Build DATA_MSG from first message in pending transmit queue using current tx sequence number. Send message to MP for transmission.
5 Dequeue first element on pending transmit queue and release its buffer Increment the tx sequence number Stop the retransmit timer if pending transmit queue not empty Build DATA_MSG from first message in pending transmit queue using current tx sequence number. Send message to MP for transmission. Reset tx retransmit count
6 Build ACK_MSG using the current rx sequence number Send ack message to MP for transmission Pass message to call managment layer Increment rx sequence number
7 Build ACK_MSG using the previous rx sequence number Send the ack message to MP for transmission
8 Add the message to the end of the pending transmit queue
The version exchange message is sent by the call originator to inform the answerer the version of the MP+ protocol being used as well as any other information that may need to be conveyed outside of the normal PPP parameter negotiation.
+---------------+---------------+ | Message type | | 0x00000001 | +---------------+---------------+ | Protocol Version | +---------------+---------------+ | Protocol Revision | +---------------+---------------+ | Session Identifier | +---------------+---------------+ | Hardware Type | +---------------+---------------+ | Nailed Mode | +---------------+---------------+ | Use Multiple Trunk Groups | +---------------+---------------+ | Descriptor Length | +---------------+---------------+ | Descriptor | +---------------+---------------+
Figure 4: Version Exchange Request
A message sent from call originator to call answerer specifying the callers protocol version and other state info and requesting the answerer to respond with its version info.
The version exchange response message is sent by the call answerer in response to a version exchange request message. The answerer uses the message to inform the caller the version of the MP+ protocol being used as well as any other information that needs to be conveyed outside of the normal PPP parameter negotiation.
+---------------+---------------+ | Message type | | 0x00000002 | +---------------+---------------+ | Protocol Version | +---------------+---------------+ | Protocol Revision | +---------------+---------------+ | Session Identifier | +---------------+---------------+ | Hardware Type | +---------------+---------------+ | Nailed Mode | +---------------+---------------+ | Use Multiple Trunk Groups | +---------------+---------------+ | Descriptor Length | +---------------+---------------+ | Descriptor | +---------------+---------------+
Figure 5: Version Exchange Response
A message sent from call answerer to the call originator specifying the answerers protocol version and other state info. Sent in response to receiving a version exchange request.
Session Identifier: A non-zero identifier unique to the answerer. 2 octets fixed length.
Smith Informational [Page 12]
RFC 1934 Multilink Protocol Plus April 1996
Hardware Type: caller hardware type (can be vendor defined). 2 octets fixed length.
Nailed Mode: caller nailed mode from the session profile. 2 octets fixed length.
Use Multiple Trunk Groups: non-zero if call may use channels from multiple trunk groups. 2 octets fixed length
Descriptor Length: length of the remote descriptor in 4-octet units. 2 octets fixed length
Descriptor: the remote unit descriptor. This field allows for vendor specific identification of the peer. Variable length Nx4 octets long - total length defined above.
A message of this type is sent by either caller or answerer to initiate an increase of bandwidth. When sent by the caller the request is asking for permission to dial a certain number of channels; the response will contain permission and the phone numbers of the channels to dial. When sent by the answerer, this message contains the phone numbers to dial. The message looks like:
+---------------+---------------+ | Message type | | 0x00000003 | +---------------+---------------+ | Number of channels requested | +---------------+---------------+ | Number of phone numbers | +---------------+---------------+ | A phone number list for | | each phone number | | . | | . | | . | +---------------+---------------+
Figure 6: Add Request
Smith Informational [Page 13]
RFC 1934 Multilink Protocol Plus April 1996
A message sent by either caller or answerer to request that additional bandwidth be added to a session.
Number of channels requested: The maximum number of channels to add. 2 octets fixed length.
Number of phone numbers: The number of phone numbers provided. This value will always be zero when the caller initiates an add and will be at least Number of channels requested when the answerer initiates the add. 2 octets fixed length.
Phone number list: A list of up to 32 phone number lists containing the phone numbers to call. Each description is of fixed length as described below:
Each phone number is represented by a phone number list. The format of a phone number list is:
+---------------+---------------+ | in use flag | +---------------+---------------+ | call service type | +---------------+---------------+ | Phone number | | 20 octets | | plus null terminator | | (21 octet total) | | | | | | | | | | | | | | +---------------+ | | Must be 0 | +---------------+---------------+ | must be 0 | +---------------+---------------+
Figure 7: Phone number list
Smith Informational [Page 14]
RFC 1934 Multilink Protocol Plus April 1996
A structure containing information about a connection within the system.
in use flag: non-zero if the phone number indicated in this descriptor is currently in use. 2 octets fixed length
call service type: Defines the type of service, switched, nailed, or other, associated with a phone number. 1 Nailed 2 Switched >=3 Undefined
Phone number: The null terminated phone number of this channel. Fixed length 21 octets. Each octet contains IA5 character representation of a digit (or #, *).
Must be 0: Filler to force alignment to 32-bit boundary.
A message of this type gives permission to dial some number of channels and, when sent by the answerer of the original call, gives the phone numbers of channels to dial.
+---------------+---------------+ | Message type | | 0x00000004 | +---------------+---------------+ | Number of channels allowed | +---------------+---------------+ | Number of phone numbers | +---------------+---------------+ | A phone number list for | | each phone number | | . | | . | | . | +---------------+---------------+
Figure 8: Add Response
Smith Informational [Page 15]
RFC 1934 Multilink Protocol Plus April 1996
A message sent by either caller or answerer to indicate the number of channels that may be added to a session.
Number of channels allowed: The actual number of channels to add. This may be less than the number requested. 2 octets fixed length.
Number of phone numbers: The number of phone numbers provided. This value will always be zero when sent by the caller and will be at least channelCount when sent by the answerer. 2 octets fixed length.
Phone number list: A list of up to 32 phone number lists containing the phone numbers to call. Each description is of fixed length as described above.
This message is sent by the caller to the answerer after all calls have been placed. The message is used to notify the answerer that the add transaction is complete and it may return to the idle state.
+---------------+---------------+ | Message type | | 0x00000005 | +---------------+---------------+ | Channels added | +---------------+---------------+ | Must be zero | +---------------+---------------+
Figure 9: Add Complete
A message sent by caller to indicate the number of channels that were added successfully. This message was added in MP+ Version 1.1.
Channels added: The actual number of channels added. 2 octets fixed length
Must be zero: Padding to 32-bit boundary. 2 octets fixed length
A message of this type is sent when a peer decides, for any reason, to remove channels from use. The purpose of the message is to tell the remote end of the remove and give it a chance to adjust the number of channels to remove.
+---------------+---------------+ | Message type | | 0x00000006 | +---------------+---------------+ | Number of channels to remove | +---------------+---------------+ | Must be zero | +---------------+---------------+
Figure 10: Remove Request
A message sent by either caller or answerer to request that bandwidth be removed from a session.
Number of channels to remove: The maximum number of channels to remove. 2 octets fixed length
Must be zero: Padding to 32-bit boundary. 2 octets fixed length
This message is sent in response to a remove request. The responder specifies the number of channels that can be removed. If the response specifies 0 channels the remove is cancelled.
+---------------+---------------+ | Message type | | 0x00000007 | +---------------+---------------+ | Number of channels to remove | +---------------+---------------+ | Must be zero | +---------------+---------------+
Figure 11: Remove Response
Smith Informational [Page 17]
RFC 1934 Multilink Protocol Plus April 1996
A message sent in response to a remove request specifying the number of channels that the peer agrees can be removed.
Number of channels to remove: The maximum number of channels to remove. May be zero, in which case the remove is cancelled. 2 octets fixed length
Must be zero: Padding to 32-bit boundary. 2 octets fixed length
This message is sent by the initiator of a remove transaction when the agreed upon number of channels have been removed. The message is used to notify the peer that the remove transaction is complete and it may return to the idle state.
+---------------+---------------+ | Message type | | 0x00000008 | +---------------+---------------+ | Number of channels removed | +---------------+---------------+ | Must be zero | +---------------+---------------+
Figure 12: Remove Complete
A message sent by the caller or answerer to indicate how many channels were actually removed. This message was added in MP+ CM version 1.1.
Number of channels removed: The number of channels that were removed. 2 octets fixed length
Must be zero: Padding to 32-bit boundary. 2 octets fixed length
This message is sent when the peer requests to close the whole session. This is typically due to a configuration option that indicates when a system should request to close the session (an example being, a link has been idle for greater than a preconfigured time period).
+---------------+---------------+ | Message type | | 0x00000009 | +---------------+---------------+
Figure 13: MP+ close request.
There are no data fields associated with this message.
If the peer agrees that closing the session is acceptable based on it's own configuration (an example reject reason would be that the peer is configured with a *minimum* number of channels to keep active).
+---------------+---------------+ | Message type | | 0x0000000a | +---------------+---------------+ | OK To Close | +---------------+---------------+ | Must be zero | +---------------+---------------+
Figure 14: MP+ close response
The response to a close request. May be sent by caller or answerer.
OK To Close: If non-zero, peer said I can close all channels. 2 octets fixed length
Must be zero: Padding to 32-bit boundary. 2 octets fixed length
This message is sent from a master station to a slave station when the master wishes to manage the remote station. The message is also used to cancel remote management once it's been started.
+---------------+---------------+ | Message type | | 0x0000000b | +---------------+---------------+ | Mode | +---------------+---------------+ | Must be zero | +---------------+---------------+
Figure 15: Remote Management Request
A message sent from master to slave to initiate or clear a remote management session.
Mode: One to start session. Zero to stop session. 2 octets fixed length
Must be zero: Padding to 32-bit boundary. 2 octets fixed length
The slave side of a remote management session has the opportunity to reject remote management. The master side is informed of accept/deny status via the remote management response.
This message type is used to convey keyboard input from the management master to be processed by the management slave. The message format consists of an octet count (in network byte order) and then an array of octets to be processed. It looks like:
+---------------+---------------+ | Message type | | 0x0000000d | +---------------+---------------+ | character count | +---------------+---------------+ | array of characters | | . | | . | | . | +---------------+---------------+
Figure 17: Remote Management Receive Request
A message sent from master to slave, conveying keystrokes typed on the masters keyboard that will be processed by the slave.
character count: Number of characters to process. 2 octets fixed length
array of characters: Array of characters to process.
The remote management slave conveys output to be displayed on the masters terminal with a remote management transmit request message. Only one message may be outstanding. The next transmit request may not be sent until the previous has been acknowledged.
+---------------+---------------+ | Message type | | 0x0000000e | +---------------+---------------+ | character count | +---------------+---------------+ | array of characters | | . | | . | | . | +---------------+---------------+
Figure 18: Remote Management Transmit Request
A message sent from slave to master, conveying output to be output on the master's display.
Character count: Number of characters to process. 2 octets fixed length
array of characters: Array of characters to process.
This message is used to acknowledge remote management transmit requests. The slave may send the next transmit request once this message has been received.
+---------------+---------------+ | Message type | | 0x0000000f | +---------------+---------------+
Figure 19: Remote Management Transmit Response
There are no data fields associated with this message.
A message sent to initiate a friendly shutdown of an MP+ link. The sender will stop sending data immediately. The receiver of the message will also stop sending user data and start a clean shutdown of all NCPs and the LCP of each member link of the bundle. When the last member link terminates, the session is completely closed.
+---------------+---------------+ | Message type | | 0x00000010 | +---------------+---------------+
Figure 20: Clear Request
There are no data fields associated with this message.
The MP+ state machine is event driven. Reception of an event triggers an action and possibly a state change. The events processed by the MP+ state machine can be roughly classed into two types:
Events that originate within the unit, e.g. notification that a call has cleared, an MP+ session may be started, etc.
Events that originate with the reception of an MP+ control message from the peer unit.
Both types are processed by the state machine in the sequence they occurred. The events processed are:
MP+START_SESSION: Notification from PPP/MP that an MP+ session is starting.
MP+SESSION_DOWN: Notification from the error-control layer that end-to-end connectivity has been lost and control messages can not be delivered.
MP+SESSION_TERM: Session termination notification from PPP/MP. This event is not sent until the last channel of a multi-channel session is cleared.
MP+TIMER_EXPIRED: Timers are used in various states and sub-states. This event is signaled whenever a timer expires.
Smith Informational [Page 23]
RFC 1934 Multilink Protocol Plus April 1996
MP+CALL_COMPLETE: A call placed during an add request has completed. The call may have succeeded or failed.
MP+UTILIZATION: Notification from MP/PPP that link utilization has crossed a threshold and that channels may need to be added/removed. (The number of channels to add/remove will be passed with the notification).
MP+RX_VERSION_REQ: A Version Exchange request message has been received from the peer.
MP+RX_VERSION_RSP: A Version Exchange response message has been received from the peer.
MP+ADD_REQ: An Add request message has been received from the peer.
MP+ADD_RSP: An Add response message has been received from the peer.
MP+ADD_COMP: An Add Complete message has been received from the peer.
MP+REMOVE_REQ: A Remove request message has been received from the peer.
MP+REMOVE_RSP: A Remove response message has been received from the peer.
MP+REMOVE_COMP: A Remove Complete message has been received from the peer.
MP+RX_RM_REQ: A Remote Management request has been received from the peer.
MP+RX_RM_RSP: A Remote Management response has been received from the peer.
MP+RX_RM_RX_REQ: A Remote Management Receive Request has been received from the far end.
MP+RX_RM_TX_REQ: A Remote Management Transmit Request has been received from the peer.
Smith Informational [Page 24]
RFC 1934 Multilink Protocol Plus April 1996
MP+RX_RM_TX_RSP: A Remote Management Transmit Response has been received from the peer.
MP+RX_CLEAR: A request to shut down the session has been received from the peer.
MP+CLOSE_REQ: A Close Request message has been received from the peer.
MP+CLOSE_RSP: A Close Response message has been received from the peer.
MP+START_RM Request to start a remote management session with this station being the master.
MP+SEND_RMS: Request to send data to a remote management master from a slave.
MP+SEND_RMM: Request to send data to a remote management slave from a master.
MP+RECV_RMM: Request to send an ack to a remote management slave for data received from the slave.
MP+STOP_RM: Request to stop a remote management session.
To ease readability and understanding the major states are considered as separate state machines, each having two to four sub-states. The sub-states are named by the letters, A, B, C, and D. State information is maintained for every interface.
The major states are:
MP+STATE_INITIAL: The state of an unused session. Session table entries are intialized to this state at startup and return to this state when sessions are terminated.
MP+STATE_IDLE: The state of an active session that is not performing any MP+ function.
MP+STATE_ADD: The state of a session when an add transaction is in progress.
Smith Informational [Page 25]
RFC 1934 Multilink Protocol Plus April 1996
MP+STATE_REMOVE: The state of a session when a remove transaction is in progress.
MP+STATE_CLOSE: The state of a session that is in the process of being closed.
State transitions are triggered by the reception of an event. Tables 2 through 6 contain the state tables for the major states. All state tables use the following symbols.
- No action
* Stay in same state
+ Target state is defined by the action taken
** An error has occurred, log an error message but no state change.
States and sub-state transitions are noted as state:sub-state, e.g., initial:A. Alternative transitions are listed on separate lines.
Some actions are common to all states, they are defined here.
Error Close Action
Called to close a session when an error occurs. Actions are:
[1] Stop timer if running.
[2] Log an error message.
[3] Close the MP+EC layer for this session.
[4] Close MP for this session
[5] Clean up, restore state variables to their initial state.
Term Action
Processed when a MP+SESSION_TERM event occurs in most states. Actions are:
[1] Stop timer if running.
[2] Close the MP+EC layer for this session.
Smith Informational [Page 26]
RFC 1934 Multilink Protocol Plus April 1996
[3] Call the passed termination callback function if not null.
[4] Clean up, restore state variables to their initial state.
Ignore Action
We don't care about this event in this state. Do nothing.
Timer Action
This action is called when a timer expires in one of the on-line states. The timer is used to implement add and remove locks. A lock is set when an add or remove fails and is not cleared until a bandwidth change or the timer expires. This keeps us from retrying add's and subtracts until there is a likelyhood that it will succeed.
[1] Check add lock flag.
[1] If set an add lock occured last timeout period so triple the timeout value (to a max of 81 minutes).
[2] If not set restore the timeout value to its initial value of one minute.
[2] Clear the add lock flag.
[3] Clear the remove lock flag.
[4] Restart the retry timer.
Enter Remove [local] Action
The local unit is initiating a remove transaction. The desired bandwidth is given.
[1] Restart the idle timer.
[2] Calculate number of channels to remove (difference between number in use and number in desired).
[3] Build and send a remove request and send to remote.
[4] Go to REMOVE:A.
Smith Informational [Page 27]
RFC 1934 Multilink Protocol Plus April 1996
Enter Remove [remote] Action
The remote unit is initiating a remove transaction. The incoming message contains the number of channels to remove.
[1] Restart the idle timer.
[2] Request the number of channels required. If greater than the number available after removing the number of channels indicated in the incoming message reduce the number of channels to remove and set a remove lock.
[3] Build a remove response message indicating the number of channels we will allow the requester to remove and send to the remote.
[4] Go to REMOVE:B.
Enter Add [local] Action
The local unit is initiating an add transaction. We are given the number of channels desired. The steps are:
[1] Restart the idle timer.
[2] Calculate number of channels to add (difference between number desired and number in use).
[3] Reserve number of channels, retrieving their phone numbers.
[4] If number of channels reserved less than the number desired set an add lock.
[5] If number of channels reserved greater than zero.
[1] Build an add request. If the answerer the request includes the phone numbers for the caller to dial.
[2] If caller, go to ADD:A.
[3] If answerer, go to ADD:C.
[6] Go to IDLE state.
Smith Informational [Page 28]
RFC 1934 Multilink Protocol Plus April 1996
Enter Remove [remote] Action
The remote unit is initiating a remove transaction. The incoming message contains the number of channels to remove.
[1] Clear the remove lock.
[2] Restart the idle timer.
[3] Request the number of channels required. If greater than the number available after removing the number of channels indicated in the incoming message reduce the number of channels to remove.
[4] Build a remove response message indicating the number of channels we will allow the requester to remove and send to the remote.
[5] Go to REMOVE, sub-state B.
Enter Add [remote answerer] Action
We've received a message from the remote requesting that bandwidth be added. The message contains the number of channels to add. Since the remote is the answerer, the mes- sage also contains the phone nubmers to dial. We may dial less than the number requested.
[1] Restart idle timer.
[2] If the number of channels requested will put us over the maximum number of channels allowed for the session reduce the channel count.
[3] For each channel to add,
[1] Integrate the phone number returned from the answerer with the original phone number dialed.
[2] Request that a session be extended by dialing the integrated phone number. A callback is passed with the request so call success or fail- ure can be reported back to MP+.
[4] Go to ADD:B. Note: This change must actu- ally occur before requesting the first outgoing call. If not, the callback could be called (and ignored) because the session is not in the correct state.
Smith Informational [Page 29]
RFC 1934 Multilink Protocol Plus April 1996
Enter Add [remote caller] Action
We've received a message from the remote requesting that bandwidth be added. The message contains the number of channels to add. Since the remote is the caller, it needs us to send the phone numbers to dial. We may send fewer phone numbers than requested
[1] Restart idle timer.
[2] If the number of channels requested will put us over the maximum number of channels allowed for the session reduce the channel count.
[3] Reserve the adjusted number of channels, retrieving their phone numbers.
[4] If the number of channels reserved is less than the adjusted number requested.
[5] Build an add response message, including the phone numbers for the channels we will let the caller add and send it to the far end.
[6] Go to ADD:C.
Enter Idle Action
The IDLE state is entered at the end of normal transactions. At entry the current status of the connection should be checked and new transactions initiated if necessary. To be safe, we can also use this state as a catch all place to release any bandwidth reserved for adds. The functions to perform are:
[1] Restart the idle timer using the current retry value.
[2] Release any reserved bandwidth not actually in use.
[3] Check if bandwidth change reqested during last trans- action. If change indicated:
[1] Query channel counts.
[2] If current bandwidth less than suggested band- width and removes are not locked, store the requested bandwidth and initiate a remove trans- action (Enter Remove Action).
Smith Informational [Page 30]
RFC 1934 Multilink Protocol Plus April 1996
[3] If current bandwidth greater than suggested bandwidth and adds are not locked:
[1] Store the requested bandwidth.
[2] Intiate an add transaction (Enter Add [local] Action).
[4] Go to the IDLE state.
Remote Management Request Action
We received a request to start/stop remote management. If this is a start request If we can/allow remote management: Build and send a Remote management response Allow message. Else Build and send a Remote management response Deny message. Else (this is a stop) Notify the remote management slave process to terminate.
Remote Management Response Action
We received a response to our remote management start request. If the response was an Allow response Notify the remote management master process, we can start sending keystrokes/commands Else The peer denied the request, so notify the remote management master process of failure.
Remote Management Receive Data Action
We (the slave) received data from the remote management master. Pass the received data to the remote management slave process. This is typically keystroke data received from the remote user interface.
Remote Management Transmit Data Action
We (the master) received data from the remote management slave. Pass the received data to the remote management master process. This is typically screen-updates to be passed to the user interface.
Smith Informational [Page 31]
RFC 1934 Multilink Protocol Plus April 1996
Remote Management Transmit Data Response Action
We (the slave) received an ack to data we previously sent to the master. Notify the remote management slave process so that it can queue further transmissions.
Remote management (Master) start Action
Build a REMOTE_MGMT_REQ start message and send to the far end. Send a proceeding message to the RM master process.
Remote management (Slave) data Action
Build a REMOTE_MGMT_TX_REQ message with the data passed from the remote management slave process, send it to the far end.
Remote management (Master) data Action
Build a REMOTE_MGMT_RX_REQ message with data passed from the remote management master process, send it to the far end.
Remote management data acknowledgement Action
Build a REMOTE_MGMT_TX_RSP message and send it so the slave can send the next block. There is no data associated with this message.
Remote management (Master) stop Action
Build a REMOTE_MGMT_REQ stop message and send to the far end.
Compare protocol versions, if they do not match Do Error Close Action, go to INITIAL, sub-state A.
Store off info received from remote.
Check the base channel count in the callers profile.
If greater than 1
Set the requested bandwidth to the base channel count.
Do Enter Add Caller action which causes a state change.
Do Enter Idle Action which causes a state change.
7 Both sides think they are the answerer. This is possible if both dial each other at the same time and the first channel that completed PPP negotiation happened to be the channel associated with the incoming call on both units. We resolve this by trying to become the originator. If both sides try to become the originator the one with the largest endpoint discriminator will fall back to being the answerer.
Restart Idle timer at 60 seconds
Build and send Version Exchange Request message
Go to Initial:B
8 Both sides think they are the originator. This can happen if both dial each other at the same time and the first channel that completed PPP negotiation happened to be the channel associated with the originating call on both units. MP+ determines which will be the caller and which the answerer by comparing the endpoind discriminator in the version exchange request with the local endpoint discriminator. The unit with the smaller endpoint is arbitrarily called the originator. The actions are:
Smith Informational [Page 34]
RFC 1934 Multilink Protocol Plus April 1996
Compare local endpoint discriminator with endpoint discrimator in message.
If local endpoint discriminator is less than the remote value we are the caller, ignore the incoming message.
Otherwise, if local endpoint discriminator is greater than the remote value we are the answerer:
Compare protocol versions, if they do not match Do Error Close Action, go to INITIAL, sub-state A.
Store off info received from remote.
Build a version exchange response and send to remote end.
Do Enter Idle Action which causes a state change.
If the two values match, there is a problem, Do Error Close Action,go to INITIAL, sub-state A.
9 Log an error message. Notify the user interface of remote management failure.
11 Do Remote Management Transmit Data Response Action
12 Clear remove lock. If local recommended channels == 0, then: send a Close Response message with OK To Close set to TRUE. Else send a Close Response message with OK To Close set to FALSE. Do Enter Idle Action.
The add state is used by both caller and answerer when an add transaction is in progress.
The sub-states are:
A Add request sent to answerer, waiting for add response from the answerer. B Caller waiting for call complete notification for calls placed. C Answerer waiting for add complete message from caller.
1 Phone numbers (may) have been reserved, they must be released before the normal error processing occurs.
Release all reserved phone numbers
Do Error Close Action.
2 Phone nubmers (may) have been reserved, they must be released before the normal close processing occurs.
Release all reserved phone numbers
Do Term Action.
3 Do Timer Action
4 Note that a bandwidth change has been reqested. This will be processed the next time IDLE state is entered.
5 An add collision has occured. Since the answerer has sent phone numbers we will try to use what he as sent, within the limits of the local system.
Compare local channels to add with current channels to add.
If the local channels to add is less than the remote channels to add
If the remote number of channels requested will put us over the maximum number of channels allowed for the session reduce the channel count and set an add lock.
Re-reserve the channels. If the number reserved are less than the number of phone numbers provided by the far end, set an add lock and reduce the number of channels to add to what we could reserve.
Now treat the remote add request as if it were an add response and process by:
Integrate the phone number returned from the answerer with the original phone number dialed.
Smith Informational [Page 39]
RFC 1934 Multilink Protocol Plus April 1996
Request that a session be extended by dialing the integrated phone number. A callback is passed with the request so call success or failure can be reported back to MP+.
Go to ADD:B. Note: This change must actually occur before requesting the first outgoing call. If not, the callback could be called (and ignored) because the session is not in the correct state.
6 If the answerer provided fewer phone numbers than requested set an add lock.
If the number of channels is zero send an add complete message (there's nothing to do) and go to the IDLE state.
For each phone number returned
Integrate the phone number returned from the answerer with the original phone number dialed.
Request that a session be extended by dialing the integrated phone number. A callback is passed with the request so call success or failure can be reported back to MP+.
Go to ADD:B. Note: This change must actually occur before requesting the first outgoing call. If not, the callback could be called (and ignored) because the session is not in the correct state.
7 Restart idle timer for abort.
8 Increment the count of calls completed.
If the call succeeded, increment the count of calls that succeeded.
If the count of calls completed equals the number of calls placed
If number of calls completed is not the same as the nubmer that succeeded set an add lock.
Build an add complete message and send it to the far end.
If at least one channel was added clear any remove lock.
Smith Informational [Page 40]
RFC 1934 Multilink Protocol Plus April 1996
Go to the IDLE state.
9 If number of channels requested not equal to number connected set add lock.
If at least one channel was added clear any remove lock.
The close state is used when we are gracefully closing a session or when we were notified that a session terminated mid-transaction.
The sub-states are:
A Waiting for call complete after session down notification B Waiting for call complete after session terminate notification. C Waiting for close response after session close request sent.
1 The session was closed while waiting for call completes. Just go to Close:B.
2 We timed out waiting for completes. Just process the link down, now. Do Error Close Action.
3 Increment the number of calls complete.
If equal to the number of calls placed then: Do Error Close Action, go to Initial:A. Else No state change.
4 Log an error message. Notify the user interface of remote management failure.
5 We didn't get all the notifications that we expect. Give up and close the session anyway. Do Term Action .
6 Increment the number of calls complete.
If equal to the number of calls placed then: Do Term Action, go to Initial:A. Else No state change
7 Do Error Close Action
8 Do Term Action
9 Note that a bandwidth change has been reqested. This will be
Smith Informational [Page 45]
RFC 1934 Multilink Protocol Plus April 1996
processed the next time IDLE state is entered.
10 This is an Add & Close collision. Add wins. Perform current remote add action. If we are originator Do Add [Remote Answerer] Action else Do Add [Remote Caller] Action
11 This is a Remove & Close collision, the Remove will win: Set remove lock to FALSE Do Remove [Remote] Action.
12 This is a Close collision. But since we both agree: If we are originator Send a Close Response with okToClose set to TRUE. Else Send a Close Response with okToClose set to FALSE.
13 If Close Response is received with okToClear is TRUE then: Do Term Action Else set remove lock to TRUE and do Enter Idle Action.
4. PPP LCP Extensions
MP+ Configuration Option
The Multilink Protocol Plus introduces the use of an additional LCP Configuration Option: