rfc2892.txt

<VC++网络游戏建摸与实现>源代码

   stripped. Unicast packets are normally stripped by the destination
   station or by the source station if the destination station has
   failed. Multicast packets are only stripped by the source station. If
   both the source and destination stations drop out of the ring while a
   unicast packet is in flight, or if the source node drops out while
   its multicast packet is in flight, the packet will rotate around the
   ring continuously.

   The solution to this problem is to have a TTL or Time To Live field
   in each packet that is set to at least twice the number of nodes in
   the ring. As each node forwards the packet, it decrements the TTL. If
   the TTL reaches zero it is stripped off of the ring.





Tsiang & Suwala              Informational                     [Page 24]

RFC 2892            The Cisco SRP MAC Layer Protocol         August 2000


   The ring ID is used to qualify all stripping and receive decisions.
   This is necessary to handle the case where packets are being wrapped
   by some node in the ring. The sending node may see its packet on the
   reverse ring prior to reaching its destination so must not source
   strip it.  The exception is if a node is in wrap.  Logically, a node
   in wrap "sees" the packet on both rings.  However the usual
   implementation is to receive the packet on one ring and to transmit
   it on the other ring.  Therefore, a node that is in the wrap state
   ignores the ring ID when making stripping and receiving decisions.

   A potential optimization would be to allow ring ID independent
   destination stripping of unicast packets.  One problem with this is
   that packets may be delivered out of order during a transition to a
   wrap condition. For this reason, the ring ID should always be used as
   a qualifier for all strip and receive decisions.

5.  Packet acceptance and stripping

   A series of decisions based on the type of packet (mode), source and
   destination addresses are made on the MAC incoming packets. Packets
   can either be control or data packets.  Control packets are stripped
   once the information is extracted. The source and destination
   addresses are checked in the case of data packets. The rules for
   reception and stripping are given below as well as in the flow chart
   in Figure 16.

      1. Decrement TTL on receipt of a packet, discard if it gets to
         zero; do not forward.

      2. Strip unicast packets at the destination station.  Accept and
         strip "control" packets.

      3. Do not process packets other than for TTL and forwarding if
         they have the "wrong" ring_id for the direction in which they
         are received unless the node is in wrap.  If the node is in
         wrap then ignore the ring_id.

      4. Do not process packets other than for TTL and forwarding if the
         mode is not supported by the node (e.g. reserved modes, or ATM
         cell mode for packet nodes).

      5. Packets accepted by the host because of the destination address
         should be discarded at the upper level if there is CRC error.

      6. Control messages are point to point between neighbors and
         should always be accepted and stripped.





Tsiang & Suwala              Informational                     [Page 25]

RFC 2892            The Cisco SRP MAC Layer Protocol         August 2000


      7. Packets whose source address is that of the receiving station
         and whose ring_id matches should be stripped.  If a node is in
         wrap then ignore the ring_id.

   FIGURE 16. SRP Receive Flowchart (Packet node)

   if (MODE == 4,5)-------------------------------->[to host]--->|
           |                                                     |
           v                                                     |
   if (MODE == 6)---------------------------------->[strip]----->|
           |                                                     |
           v                                                     |
   if (!WRAPPED                                                  |
      & WRONG_RING_ID)-------------------------------------------|--->|
           |                                                     |    |
           v                                                     |    |
   if (MODE == 0,1,2,3)------------------------------------------|--->|
           |                                                     |    |
           v                                                     |    |
   if (DA MATCH)--------------->if !(SA MATCH)----->[to host]--->|    |
           |                            |                        |    |
           |                            v                        |    |
           |                    if (unicast)------->[to host]--->|    |
           |                            |                        |    |
           |                            v                        |    |
   if (SA MATCH)-------------------->[strip]-------------------->|    |
           |                                                     |    |
           |                                                     |    v
           |--------------------------->|<-----------------------|----|
                                        |                        |
                                        v                        |
                                if (ttl < 2)------->[strip]----->|
                                        |                        |
                                        v                        |
                                [decrement ttl]                  |
                                        |                        |
                                [fwd pkt to tb]                  |
                                        |                        v
                                        |<-----------------------|
                                        v
                                  [back to top]

   Notes:  Host is responsible for discarding CRC errored packets.
           Conditionals (if statements) branch to the right if true
           and branch down if false.






Tsiang & Suwala              Informational                     [Page 26]

RFC 2892            The Cisco SRP MAC Layer Protocol         August 2000


5.1.  Transmission and forwarding with priority

   A node can transmit four types of packets:

      1. High priority packets from the high priority transit
         buffer.

      2. Low priority packets from the low priority transit buffer.

      3. High priority packets from the host Tx high priority fifo.

      4. Low priority packets from the host Tx low priority fifo.

   High priority packets from the transit buffer are always sent first.
   High priority packets from the host are sent as long as the low
   priority transit buffer is not full.  Low priority packets are sent
   as long as the transit buffer has not crossed the low priority
   threshold and the SRP-fa rules allow it (my_usage < allowed_usage).
   If nothing else can be sent, low priority packets from the low
   priority transit buffer are sent.

   This decision tree is shown in Figure 17.





























Tsiang & Suwala              Informational                     [Page 27]

RFC 2892            The Cisco SRP MAC Layer Protocol         August 2000


   FIGURE 17. SRP transmit flowchart

   if (TB_High has pkt)----------->[send pkt from TB_high]-->|
           |                                                 |
           v                                                 |
   if (TB_Low full)------------------------------------------|---->|
           |                                                 |     |
           v                                                 |     |
   if (Tx_High has pkt)----------->[send pkt from Tx_high]-->|     |
           |                                                 |     |
           v                                                 |     |
   if (TB_Low > Hi threshold)--------------------------------|---->|
           |                                                 |     |
           v                                                 |     |
   if (my_usage >= allowed_usage)----------------------------|---->|
           |                                                 |     |
           v                                                 |     |
   if (Tx_Low has pkt)------------>[send pkt from Tx_low]--->|     |
           |                                                 |     |
           |                                                 |     v
           |<------------------------------------------------|-----|
           |                                                 |
           v                                                 |
   if (TB_Low has pkt)------------>[send pkt from TB_low]--->|
           |                                                 v
           |<------------------------------------------------|
           |
           v
       [Go to Top]

   Notes:  Conditionals (if statements) branch to the right if true
           and branch down if false.

5.2.  Wrapping of Data

   Normally, transmitted data is sent on the same ring to the downstream
   neighbor.  However, if a node is in the wrapped state, transmitted
   data is sent on the opposite ring to the upstream neighbor.

6.  SRP-fa Rules Of Operation

   The SRP-fa governs access to the ring.  The SRP-fa only applies to
   low priority traffic.  High priority traffic does not follow SRP-fa
   rules and may be transmitted at any time as long as there is
   sufficient transit buffer space.






Tsiang & Suwala              Informational                     [Page 28]

RFC 2892            The Cisco SRP MAC Layer Protocol         August 2000


   The SRP-fa requires three counters which control the traffic
   forwarded and sourced on the SRP ring. The counters are my_usage
   (tracks the amount of traffic sourced on the ring), forward_rate
   (amount of traffic forwarded on to the ring from sources other than
   the host) and allowed_usage (the current maximum transmit usage for
   that node).

   With no congestion all nodes build up allowed usage periodically.
   Each node can send up to max_usage.  Max_usage is a per node
   parameter than limits the maximum amount of low priority traffic a
   node can send.

   When a node sees congestion it starts to advertise its my_usage which
   has been low pass filtered (lp_my_usage).

   Congestion is measured by the transit buffer depth crossing a
   congestion threshold.

   A node that receives a non-null usage message (rcvd_usage) will set
   its allowed usage to the value advertised.  However, if the source of
   the rcvd_usage is the same node that received it then the rcvd_usage
   shall be treated as a null value.  When comparing the rcvd_usage
   source address the ring ID of the usage packet must match the
   receiver's ring ID in order to qualify as a valid compare.  The
   exception is if the receive node is in the wrap state in which case
   the usage packet's ring ID is ignored.

   Nodes that are not congested and that receive a non-null rcvd_usage
   generally propagate rcvd_usage to their upstream neighbor else
   propagate a null value of usage (all 1's).  The exception is when an
   opportunity for local reuse is detected. Additional spatial reuse
   (local reuse) is achieved by comparing the forwarded rate (low pass
   filtered) to allow_usage.  If the forwarded rate is less than the
   allowed usage, then a null value is propagated to the upstream
   neighbor.

   Nodes that are congested propagate the smaller of lp_my_usage and
   rcvd_usage.

   Convergence is dependent upon number of nodes and distance.
   Simulation has shown simulation convergence within 100 msec for rings
   of several hundred miles.









Tsiang & Suwala              Informational                     [Page 29]

RFC 2892            The Cisco SRP MAC Layer Protocol         August 2000


6.1.  SRP-fa pseudo-code

   A more precise definition of the fairness algorithm is shown below:

Variables:

lo_tb_depth     low priority transit buffer depth

my_usage        count of octets transmitted by host
lp_my_usage     my_usage run through a low pass filter
my_usage_ok     flag indicating that host is allowed to transmit

allow_usage     the fair amount each node is allowed to transmit

fwd_rate        count of octets forwarded from upstream
lp_fwd_rate     fwd_rate run through a low pass filter

congested       node cannot transmit host traffic without the TB buffer
                filling beyond its congestion threshold point.

rev_usage       the usage value passed along to the upstream neighbor

Constants:

MAX_ALLOWANCE = configurable value for max allowed usage for this node

DECAY_INTERVAL = 8000 octet times @ OC-12, 32,000 octet times @ OC-48

AGECOEFF = 4    // Aging coeff for my_usage and f