rfc2889.txt

来自「著名的RFC文档,其中有一些文档是已经翻译成中文的的.」· 文本 代码 · 共 1,697 行 · 第 1/5 页

Network Working Group                                      R. Mandeville
Request for Comments: 2889                                     CQOS Inc.
Category: Informational                                        J. Perser
                                                  Spirent Communications
                                                             August 2000


           Benchmarking Methodology for LAN Switching Devices

Status of this Memo

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

Copyright Notice

   Copyright (C) The Internet Society (2000).  All Rights Reserved.

Table of Contents

   1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . .  2
   2. Requirements . . . . . . . . . . . . . . . . . . . . . . . . .  2
   3. Test setup . . . . . . . . . . . . . . . . . . . . . . . . . .  2
   4. Frame formats and sizes  . . . . . . . . . . . . . . . . . . .  3
   5. Benchmarking Tests . . . . . . . . . . . . . . . . . . . . . .  3
      5.1  Fully meshed throughput, frame loss and forwarding rates   4
      5.2  Partially meshed one-to-many/many-to-one  . . . . . . . .  7
      5.3  Partially meshed multiple devices . . . . . . . . . . . . 10
      5.4  Partially meshed unidirectional traffic . . . . . . . . . 13
      5.5  Congestion Control  . . . . . . . . . . . . . . . . . . . 16
      5.6  Forward Pressure and Maximum Forwarding Rate  . . . . . . 19
      5.7  Address caching capacity  . . . . . . . . . . . . . . . . 22
      5.8  Address learning rate . . . . . . . . . . . . . . . . . . 25
      5.9  Errored frames filtering. . . . . . . . . . . . . . . . . 27
      5.10 Broadcast frame Forwarding and Latency  . . . . . . . . . 28
   6. Security Considerations  . . . . . . . . . . . . . . . . . . . 30
   7. References . . . . . . . . . . . . . . . . . . . . . . . . . . 30
   8. Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . 30
      Appendix A: Formulas . . . . . . . . . . . . . . . . . . . . . 31
      Appendix B: Generating Offered Load  . . . . . . . . . . . . . 32
      Full Copyright Statement . . . . . . . . . . . . . . . . . . . 35









Mandeville & Perser          Informational                      [Page 1]

RFC 2889          LAN Switch Benchmarking Methodology        August 2000


1. Introduction

   This document is intended to provide methodology for the benchmarking
   of local area network (LAN) switching devices.  It extends the
   methodology already defined for benchmarking network interconnecting
   devices in RFC 2544 [3] to switching devices.

   This RFC primarily deals with devices which switch frames at the
   Medium Access Control (MAC) layer. It provides a methodology for
   benchmarking switching devices, forwarding performance, congestion
   control, latency, address handling and filtering. In addition to
   defining the tests, this document also describes specific formats for
   reporting the results of the tests.

   A previous document, "Benchmarking Terminology for LAN Switching
   Devices" [2], defined many of the terms that are used in this
   document.  The terminology document SHOULD be consulted before
   attempting to make use of this document.

2. Requirements

   The following RFCs SHOULD be consulted before attempting to make use
   of this document: RFC 1242 [1], RFC 2285 [2], and RFC 2544 [3].

   For the sake of clarity and continuity, this RFC adopts the template
   for benchmarking tests set out in Section 26 of RFC 2544.

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

3. Test setup

   This document extends the general test setup described in section 6
   of RFC 2544 [3] to the benchmarking of LAN switching devices.  RFC
   2544 [3] primarily describes non-meshed traffic where input and
   output interfaces  are grouped in mutually exclusive sending and
   receiving pairs.  In fully meshed traffic, each interface of a
   DUT/SUT is set up to both receive and transmit frames to all the
   other interfaces under test.

   Prior to each test run, the DUT/SUT MUST learn the MAC addresses used
   in the test and the address learning SHOULD be verified.  Addresses
   not learned will be forwarded as flooded frames and reduce the amount
   of correctly forwarded frames.  The rate at which address learning
   frames are offered may have to be adjusted to be as low as 50 frames
   per second or even less, to guarantee successful learning.  The
   DUT/SUT address aging time SHOULD be configured to be greater than



Mandeville & Perser          Informational                      [Page 2]

RFC 2889          LAN Switch Benchmarking Methodology        August 2000


   the period of the learning phase of the test plus the trial duration
   plus any configuration time required by the testing device.
   Addresses SHOULD NOT age out until the trial duration is completed.
   More than one learning trial may be needed for the association of the
   address to the port to occur.

   If a DUT/SUT uses a hashing algorithm with address learning, the
   DUT/SUT may not learn the necessary addresses to perform the tests.
   The format of the MAC addresses MUST be adjustable so that the
   address mapping may be re-arranged to ensure that the DUT/SUT learns
   all the addresses.

4.  Frame formats and sizes

   The test frame format is defined in RFC 2544 section 8 [3] and MUST
   contain a unique signature field located in the UDP DATA area of the
   Test Frame (see Appendix C [3]).  The purpose of the signature field
   is filter out frames that are not part of the offered load.

   The signature field MUST be unique enough to identify the frames not
   originating from the DUT/SUT.  The signature field SHOULD be located
   after byte 56 (collision window [4] ) or at the end of the frame. The
   length, contents and method of detection is not defined in this memo.

   The signature field MAY have a unique identifier per port.  This
   would filter out misforwarded frames.  It is possible for a DUT/SUT
   to strip off the MAC layer, send it through its switching matrix, and
   transmit it out with the correct destination MAC address but the
   wrong payload.

   For frame sizes, refer to RFC 2544, section 9 [3].

   There are three possible frame formats for layer 2 Ethernet switches:
   standard MAC Ethernet frames, standard MAC Ethernet frames with
   vendor-specific tags added to them, and IEEE 802.3ac frames tagged to
   accommodate 802.1p&Q.  The two types of tagged frames may exceed the
   standard maximum length frame of 1518 bytes, and may not be accepted
   by the interface controllers of some DUT/SUTs. It is recommended to
   check the compatibility of the DUT/SUT with tagged frames before
   testing.

   Devices switching tagged frames of over 1518 bytes will have a
   different maximum forwarding rate than untagged frames.

5. Benchmarking Tests

   The following tests offer objectives, procedures, and reporting
   formats for benchmarking LAN switching devices.



Mandeville & Perser          Informational                      [Page 3]

RFC 2889          LAN Switch Benchmarking Methodology        August 2000


5.1  Fully meshed throughput, frame loss and forwarding rates

5.1.1 Objective

   To determine the throughput, frame loss and forwarding rates of
   DUT/SUTs offered fully meshed traffic as defined in RFC 2285 [2].

5.1.2 Setup Parameters

   When offering full meshed traffic, the following parameters MUST be
   defined.  Each parameter is configured with the following
   considerations.

      Frame Size - Recommended frame sizes are 64, 128, 256, 512, 1024,
      1280 and 1518 bytes, per RFC 2544 section 9 [3].  The four CRC
      bytes are included in the frame size specified.

      Interframe Gap (IFG) - The IFG between frames inside a burst MUST
      be at the minimum specified by the standard (9.6 us for 10Mbps
      Ethernet, 960 ns for 100Mbps Ethernet, and 96 ns for 1 Gbps
      Ethernet) of the medium being tested.

      Duplex mode - Half duplex or full duplex.

      ILoad - Intended Load per port is expressed in a percentage of the
      medium's maximum theoretical load, regardless of traffic
      orientation or duplex mode.  Certain test configurations will
      theoretically over-subscribe the DUT/SUT.

      In half duplex, an ILoad over 50% will over-subscribe the DUT/SUT.

      Burst Size - The burst size defines the number of frames sent
      back-to-back at the minimum legal IFG [4] before pausing
      transmission to receive frames.  Burst sizes SHOULD vary between 1
      and 930 frames.  A burst size of 1 will simulate constant load
      [1].

      Addresses per port - Represents the number of addresses which are
      being tested for each port.  Number of addresses SHOULD be a
      binary exponential (i.e. 1, 2, 4, 8, 16, 32, 64, 128, 256, ...).
      Recommended value is 1.

      Trial Duration - The recommended Trial Duration is 30 seconds.
      Trial duration SHOULD be adjustable between 1 and 300 seconds.







Mandeville & Perser          Informational                      [Page 4]

RFC 2889          LAN Switch Benchmarking Methodology        August 2000


5.1.3 Procedure

   All ports on the tester MUST transmit test frames either in a Frame
   Based or Time Based mode (Appendix B).  All ports SHOULD start
   transmitting their frames within 1% of the trial duration.  For a
   trial duration of 30 seconds, all ports SHOULD have started
   transmitting frames within 300 milliseconds of each other.

   Each port in the test MUST send test frames to all other ports in a
   round robin type fashion.  The sequence of addresses MUST NOT change
   when congestion control is applied.  The following table shows how
   each port in a test MUST transmit test frames to all other ports in
   the test.  In this example, there are six ports with 1 address per
   port:

   Source Port       Destination Ports (in order of transmission)

   Port #1           2       3       4       5       6       2...
   Port #2           3       4       5       6       1       3...
   Port #3           4       5       6       1       2       4...
   Port #4           5       6       1       2       3       5...
   Port #5           6       1       2       3       4       6...
   Port #6           1       2       3       4       5       1...

   As shown in the table, there is an equal distribution of destination
   addresses for each transmit opportunity. This keeps the test balanced
   so that one destination port is not overloaded by the test algorithm
   and all ports are equally and fully loaded throughout the test.  Not
   following this algorithm exactly will produce inconsistent results.

   For tests using multiple addresses per port, the actual port
   destinations are the same as described above and the actual
   source/destination address pairs SHOULD be chosen randomly to
   exercise the DUT/SUT's ability to perform address lookups.

   For every address, learning frames MUST be sent to the DUT/SUT to
   allow the DUT/SUT update its address tables properly.

5.1.4 Measurements

   Each port should receive the same number of test frames that it
   transmitted.  Each receiving port MUST categorize, then count the
   frames into one of two groups:

      1.) Received Frames: received frames MUST have the correct
          destination MAC address and SHOULD match a signature field.

      2.) Flood count [2].



Mandeville & Perser          Informational                      [Page 5]

RFC 2889          LAN Switch Benchmarking Methodology        August 2000


   Any frame originating from the DUT/SUT (spanning tree, SNMP, RIP,
    ...) MUST not be counted as a received frame.  Frames originating
   from the DUT/SUT MAY be counted as flooded frames or not counted at
   all.

   Frame loss rate of the DUT/SUT SHOULD be reported as defined in
   section 26.3 [3] with the following notes: Frame loss rate SHOULD be
   measured at the end of the trail duration.  The term "rate", for this
   measurement only, does not imply the units in the fashion of "per
   second."

5.1.4.1 Throughput

   Throughput measurement is defined in section 26.1 [3].  A search
   algorithm is employed to find the maximum Oload [2] with a zero Frame
   loss rate [1].  The algorithm MUST adjust Iload to find the
   throughput.

5.1.4.2 Forwarding Rate

   Forwarding rate (FR) of the DUT/SUT SHOULD be reported as the number
   of test frames per second that the device is observed to successfully
   forward to the correct destination interface in response to a
   specified Oload.  The Oload MUST also be cited.

   Forwarding rate at maximum offered load (FRMOL) MUST be reported as
   the number of test frames per second that a device can successfully
   transmit to the correct destination interface in response to the MOL
   as defined in section 3.6 [2]. The MOL MUST also be cited.

   Maximum forwarding rate (MFR) MUST be reported as the highest
   forwarding rate of a DUT/SUT taken from an iterative set of
   forwarding rate measurements.  The iterative set of forwarding rate
   measurements are made by adjusting Iload.  The Oload applied to the
   device MUST also be cited.

5.1.5 Reporting format

   The results for these tests SHOULD be reported in the form of a
   graph.  The x coordinate SHOULD be the frame size, the y coordinate
   SHOULD be the test results.  There SHOULD be at least two lines on
   the graph, one plotting the theoretical and one plotting the test
   results.

   To measure the DUT/SUT's ability to switch traffic while performing
   many different address lookups, the number of addresses per port MAY
   be increased in a series of tests.




Mandeville & Perser          Informational                      [Page 6]

RFC 2889          LAN Switch Benchmarking Methodology        August 2000