rfc2121.txt

来自「RFC 的详细文档！」· 文本 代码 · 共 676 行 · 第 1/2 页

Network Working Group                                      G. Armitage
Request for Comments: 2121                                    Bellcore
Category: Informational                                     March 1997


                   Issues affecting MARS Cluster Size

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

   IP multicast over ATM currently uses the MARS model [1] to manage the
   use of ATM pt-mpt SVCs for IP multicast packet forwarding. The scope
   of any given MARS services is the MARS Cluster - typically the same
   as an IPv4 Logical IP Subnet (LIS). Current IP/ATM networks are
   usually architected with unicast routing and forwarding issues
   dictating the sizes of individual LISes. However, as IP multicast is
   deployed as a service, the size of a LIS will only be as big as a
   MARS Cluster can be. This document provides a qualitative look at the
   issues constraining a MARS Cluster's size, including the impact of VC
   limits in switches and NICs, geographical distribution of cluster
   members, and the use of VC Mesh or MCS modes to support multicast
   groups.

1. Introduction

   A MARS Cluster is the set of IP/ATM interfaces that are willing to
   engage in direct, ATM level pt-mpt SVCs to perform IP multicast
   packet forwarding [1]. Each IP/ATM interface (a MARS Client) must
   keep state information regarding the ATM addresses of each leaf node
   (recipient) of each  pt-mpt SVC it has open. In  addition, each MARS
   Client receives MARS_JOIN and MARS_LEAVE messages from the MARS
   whenever there is a requirement that Clients around the Cluster need
   to update their pt-mpt SVCs for a given IP multicast group.

   The definition of Cluster 'size' can mean two things - the number of
   MARS Clients using a given MARS, and the geographic distribution of
   MARS Clients. The number of MARS Clients in a Cluster impacts on the
   amount of state information any given client may need to store while
   managing  outgoing  pt- mpt SVCs. It also impacts on the average rate
   of JOIN/LEAVE traffic that is propagated by the MARS on
   ClusterControlVC, and the number of pt-mpt VCs that may need
   modification each time a MARS_JOIN or MARS_LEAVE appears on
   ClusterControlVC.



Armitage                     Informational                      [Page 1]

RFC 2121           Issues affecting MARS Cluster Size         March 1997


   The geographic distribution of clients affects the latency between a
   client issuing a MARS_JOIN, and it finally being added onto the pt-
   mpt VCs of the other MARS Clients transmitting to the specified
   multicast group. (This latency is made up of both the time to
   propagate the MARS_JOIN, and the delay in the underlying ATM cloud's
   reaction to the subsequent ADD_PARTY messages.)

   When architecting an IP/ATM network it is important to understand the
   worst case scaling limits applicable to your Clusters. This document
   provides a primarily qualitative look at the design choices that
   impose the most dramatic constraints on Cluster size. Since the focus
   is on worst-case scenarios, most of the analysis will assume
   multicast groups that are VC Mesh based and have all cluster members
   as sources and receivers. Engineering using the worst-case boundary
   conditions, then applying optimisations such as Multicast Servers
   (MCS), provides the Cluster with a margin of safety.  It is hoped
   that more detailed quantitative analysis of Cluster sizing limits
   will be prompted by this document.

   Section 2 comments on the VC state requirements of the MARS model,
   while Sections 3 and 4 identify the group change processing load and
   latency characteristics of a cluster as a function of its size.
   Section 5 looks at how Multicast Routers (both conventional and
   combination router/switch architectures) increase the scale of a
   multicast capable IP/ATM network. Finally, Section 6 discusses how
   the use of Multicast Servers (MCS) might impact on the worst case
   Cluster size limits.


2. VC state limitations.

   Two characteristics of ATM NICs and switches will limit the number of
   members a Cluster may contain. They are:

      The maximum number of VCs that can be originated from, or
      terminate on, a port (VCmax).

      The maximum number of leaf nodes supportable by a root node
      (LEAFmax).

   We'll assume that the MARS node has similar VCmax and LEAFmax values
   as Cluster members.  VCmax affects the Cluster size because of the
   following:

      The MARS terminates a pt-pt control VC from each cluster member,
      and originates a VC for ClusterControlVC and ServerControlVC.





Armitage                     Informational                      [Page 2]

RFC 2121           Issues affecting MARS Cluster Size         March 1997


      When a multicast group is VC Mesh based, a group member terminates
      a VC from every sender to the group, per group.

      When a multicast group is MCS based, the MCS terminates a VC from
      every sender to the group.

   LEAFmax affects the Cluster size because of the following:

      ClusterControlVC from the MARS. It has a leaf node per cluster
      member (MARS Client).

      Packet forwarding SVCs out of each MARS Client for each IP
      multicast group being sent to. It has a leaf node for each group
      member when a group is VC Mesh based.

      Packet forwarding SVCs out of each MCS for each IP multicast group
      being sent to. It has a leaf node for each group member when a
      group is MCS based.

   If we have N cluster members, and M multicast groups active (using VC
   Mesh mode, and densely populated - all receivers are senders), the
   following observations may be made:

      ClusterControlVC has N leaf nodes, so
            N <= LEAFmax.

      The MARS terminates a pt-pt VC from each cluster member, and
      originates ClusterControlVC and ServerControlVC, so
            (N+2) <= VCmax.

      Each Cluster Member sources 1 VC per group, terminates (N-1) VC
      per group, originates a pt-pt VC to the MARS, and terminates 1 VC
      as a leaf on ClusterControlVC, so
            (M*N) + 2 <= VCmax.

      The VC sourced by each Cluster member per group goes to all other
      cluster members, so
            (N-1) <= LEAFmax.













Armitage                     Informational                      [Page 3]

RFC 2121           Issues affecting MARS Cluster Size         March 1997


   Since all the above conditions must be simultaneously true, we can
   see that the most constraining requirement is either:

      (M*N) + 2 <= VCmax.

   or

      N <= LEAFmax.

   The limit involving VCmax is fundamentally controlled by the VC
   consumption of group members using a VC Mesh for data forwarding,
   rather than the termination of pt-pt control VCs on the MARS. (It is
   in practice going to be very dependent on the multicast group
   membership distributions within the cluster.)

   The LEAFmax limit comes from ClusterControlVC, and is independent of
   the density of group members (or the ratios of senders to receivers)
   for active multicast groups within the cluster.

   Under UNI 3.0/3.1 the most obvious limit on LEAFmax is 2^15 (the leaf
   node ID is 15 bits wide).  However, the signaling driver software for
   most ATM NICs may impose a limit much lower than this - a function of
   how much per-leaf node state information they need to store (and are
   capable of storing) for pt-mpt SVCs.

   VCmax is constrained by the ATM NIC hardware (for available
   segmentation or reassembly instances), or by the VC capacity of the
   switch port that the NIC is attached to.  VCmax will be the smaller
   of the two.

   A MARS Client may impose its own state storage limitations, such that
   the combined memory consumption of a MARS Client and the ATM NIC's
   driver in a given host limits both LEAFmax and VCmax to values lower
   than the ATM NIC alone might have been able to support.

   It may be possible to work around LEAFmax limits by distributing the
   leaf nodes across multiple pt-mpt SVCs operating in parallel.
   However, such an approach requires further study, and doesn't solve
   the VCmax limitation associated with a node terminating too many VCs.

   A related observation can also be made that the number of MARS
   Clients in a Cluster may be limited by the memory constraints of the
   MARS itself. It is required to keep state on all the groups that
   every one of its MARS Clients have joined. For a given memory limit,
   the maximum number of MARS Clients must drop if the average number of
   groups joined per Client rises. Depending on the level of group
   memberships, this limitation may be more severe than LEAFmax.




Armitage                     Informational                      [Page 4]

RFC 2121           Issues affecting MARS Cluster Size         March 1997


3. Signaling load.

   In any given cluster there will be an 'ambient' level of
   MARS_JOIN/LEAVE activity. The dynamic characteristics of this
   activity will depend on the types of multicast applications running
   within the cluster. For a constant relative distribution of multicast
   applications we can assume that, as the number of MARS Clients in a
   given cluster rises, so does the ambient level of MARS_JOIN/LEAVE
   activity. This increases the average frequency with which the MARS
   processes and propagates MARS_JOIN/LEAVE messages.

   The existence of MARS_JOIN/LEAVE traffic also has a consequential
   impact on signaling activity at the ATM level (across the UNI and
   {P}NNI boundaries). For groups that are VC Mesh supported, each
   MARS_JOIN or MARS_LEAVE propagated on ClusterControlVC will result in
   an ADD_PARTY or DROP_PARTY message sent across the UNIs of all MARS
   Clients that are transmitting to a given group. As a cluster's
   membership increases, so does the average number of MARS Clients that
   trigger ATM signaling activity in response to MARS_JOIN/LEAVEs.

   The size of a cluster needs to be chosen to provide some level of
   containment to this ambient level of MARS and UNI/NNI signaling.

   Some refinements to the MARS Client behaviour may also be explored to
   smooth out UNI signaling transients. MARS Clients are currently
   required to initiate revalidation of group memberships only when the
   Client next sends a packet to an invalidated group SVC. A Client
   could apply a similar algorithm to decide when it should issue
   ADD_PARTYs. For example, after seeing a MARS_JOIN, wait until it
   actually has a packet to send, send the packet, then initiate the
   ADD_PARTY. As a result actively transmitting Clients would update
   their SVCs sooner than intermittently transmitting Clients.


4. Group change latencies

   The group change latency can be defined as the time it takes for all
   the senders to a group to have correctly updated their forwarding
   SVCs after a MARS_JOIN or MARS_LEAVE is received from the MARS. This
   is affected by both the number of Cluster members and the
   geographical distribution of Cluster members. (Groups that are MCS
   based create the lowest impact when new members join or leave, since
   only the MCS needs to update its forwarding SVC.) Under some
   circumstances, especially modelling or simulation environments, group
   change latencies within a cluster may be an important characteristic
   to control.





Armitage                     Informational                      [Page 5]

RFC 2121           Issues affecting MARS Cluster Size         March 1997


   As noted in the previous section, the ADD_PARTY/DROP_PARTY signaling
   load created by membership changes in VC Mesh based groups goes up as
   the number of cluster members rises (assuming worst case scenario of
   each cluster member being a sender to the group). As the UNI load
   rises, the ATM network itself may start delivering slower processing
   of the requested events.

   Wide geographic distribution of Cluster members also delays the
   propagation of MARS_JOIN/LEAVE and ATM UNI/NNI messages. The further
   apart various members are, the longer it takes for them to receive
   MARS_JOIN/LEAVE traffic on ClusterControlVC, and the longer it takes
   for the ATM network to react to ADD_PARTY and DROP_PARTY requests. If
   the long distance paths are populated by many ATM switches,
   propagation delays due to per-switch processing will add
   substantially to delays due to the speed of light.

   (Unfortunately, mechanisms for smoothing out the transient ATM
   signaling load described in section 3 have a consequence of
   increasing the group change latency, since the goal is for some of
   the senders to deliberately delay updating their forwarding SVCs.
   This is an area where the system architect needs to make a
   situation-specific trade-off.)

   It is not clear what affect the internal processing of the MARS
   itself has on group change latency, and how this might be impacted by
   cluster size.  A component of the MARS processing latency will depend
   on the specific database implementation and search algorithms as much
   as on the number of group members for the group being modified at any
   instant. Since the maximum number of group members for a given group
   is equal to the number of cluster members, there will be an indirect
   (even if small) relationship between worst case MARS processing
   latencies and cluster size.


5. Large IP/ATM networks using Mrouters

   Building a large scale, multicast capable IP over ATM network is a
   tradeoff between Cluster sizes and numbers of Mrouters. For a given
   number of hosts, the number of clusters goes up as individual
   clusters shrink. Since Mrouters are the topological intersections
   between clusters, the number of Mrouters rises as the size of
   individual clusters shrinks. (The actual number of Mrouters depends
   largely on the logical IP topology you choose to implement, since a
   single physical Mrouter may interconnect more than two Clusters at
   once.) It is a local deployment question as to what the optimal mix
   of Clusters and Mrouters will be.





Armitage                     Informational                      [Page 6]