Internet DRAFT - draft-bmwg-dcbench-methodology

draft-bmwg-dcbench-methodology



 



Internet Engineering Task Force                               L. Avramov
Internet-Draft, Intended status: Informational             Cisco Systems
Expires April 2015                                               J. Rapp
October 17, 2014                                         Hewlett-Packard






                  Data Center Benchmarking Methodology
                   draft-bmwg-dcbench-methodology-03

Abstract

   The purpose of this informational document is to establish test and
   evaluation methodology and measurement techniques for physical
   network equipment in the data center. 

Status of this Memo

   This Internet-Draft is submitted in full conformance with the
   provisions of BCP 78 and BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-
   Drafts.

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   The list of current Internet-Drafts can be accessed at
   http://www.ietf.org/1id-abstracts.html

   The list of Internet-Draft Shadow Directories can be accessed at
   http://www.ietf.org/shadow.html


Copyright Notice

   Copyright (c) 2013 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
 


Avramov & Rapp           Expires April 20, 2015                 [Page 1]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document MUST
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.









































 


Avramov & Rapp           Expires April 20, 2015                 [Page 2]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


Table of Contents

   1.  Introduction . . . . . . . . . . . . . . . . . . . . . . . . .  3
     1.1.  Requirements Language  . . . . . . . . . . . . . . . . . .  5
     1.2. Methodology format and repeatability recommendation . . . .  5
   2. Line Rate Testing . . . . . . . . . . . . . . . . . . . . . . .  5
     2.1 Objective  . . . . . . . . . . . . . . . . . . . . . . . . .  5
     2.2 Methodology  . . . . . . . . . . . . . . . . . . . . . . . .  5
     2.3 Reporting Format . . . . . . . . . . . . . . . . . . . . . .  6
   3. Buffering Testing . . . . . . . . . . . . . . . . . . . . . . .  7
     3.1 Objective  . . . . . . . . . . . . . . . . . . . . . . . . .  7
     3.2 Methodology  . . . . . . . . . . . . . . . . . . . . . . . .  7
     3.3 Reporting format . . . . . . . . . . . . . . . . . . . . . . 10
   4 Microburst Testing . . . . . . . . . . . . . . . . . . . . . . . 10
     4.1 Objective  . . . . . . . . . . . . . . . . . . . . . . . . . 10
     4.2 Methodology  . . . . . . . . . . . . . . . . . . . . . . . . 10
     4.3 Reporting Format . . . . . . . . . . . . . . . . . . . . . . 11
   5. Head of Line Blocking . . . . . . . . . . . . . . . . . . . . . 11
     5.1 Objective  . . . . . . . . . . . . . . . . . . . . . . . . . 11
     5.2 Methodology  . . . . . . . . . . . . . . . . . . . . . . . . 11
     5.3 Reporting Format . . . . . . . . . . . . . . . . . . . . . . 13
   6. Incast Stateful and Stateless Traffic . . . . . . . . . . . . . 13
     6.1 Objective  . . . . . . . . . . . . . . . . . . . . . . . . . 13
     6.2 Methodology  . . . . . . . . . . . . . . . . . . . . . . . . 13
     6.3 Reporting Format . . . . . . . . . . . . . . . . . . . . . . 14
   7.  References . . . . . . . . . . . . . . . . . . . . . . . . . . 15
     7.1.  Normative References . . . . . . . . . . . . . . . . . . . 16
     7.2.  Informative References . . . . . . . . . . . . . . . . . . 16
     7.3.  URL References . . . . . . . . . . . . . . . . . . . . . . 16
   Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 16



1.  Introduction

   Traffic patterns in the data center are not uniform and are
   constantly changing. They are dictated by the nature and variety of
   applications utilized in the data center. It can be largely east-west
   traffic flows in one data center and north-south in another, while
   some may combine both. Traffic patterns can be bursty in nature and
   contain  many-to-one, many-to-many, or one-to-many flows. Each flow
   may also be small and latency sensitive or large and throughput
   sensitive while containing a mix of UDP and TCP traffic. All of which
   can coexist in a single cluster and flow through a single network
   device all at the same time. Benchmarking of network devices have
   long used RFC1242, RFC2432, RFC2544, RFC2889 and RFC3918. These
   benchmarks have largely been focused around various latency
   attributes and max throughput of the Device Under Test [DUT] being
 


Avramov & Rapp           Expires April 20, 2015                 [Page 3]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   benchmarked. These standards are good at measuring theoretical max
   throughput, forwarding rates and latency under testing conditions
   however, they do not represent real traffic patterns that may affect
   these networking devices.

   The following provides a methodology for benchmarking Data Center DUT
   including congestion scenarios, switch buffer analysis, microburst,
   head of line blocking, while also using a wide mix of traffic
   conditions.







































 


Avramov & Rapp           Expires April 20, 2015                 [Page 4]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


1.1.  Requirements Language

   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 [6].


1.2. Methodology format and repeatability recommendation

   The format used for each section of this document is the following:

   -Objective

   -Methodology

   -Reporting Format

   MUST: minimum test for the scenario described

   SHOULD: recommended test for the scenario described

   MAY: ideal test for the scenario described

   For each test methodology described, it is key to obtain
   repeatability of the results. The recommendation is to perform enough
   iterations of the given test to make sure the result is accurate,
   this is especially important for section 3) as the buffering testing
   has been historically the least reliable. 


2. Line Rate Testing

2.1 Objective

   Provide at maximum rate test for the performance values for
   throughput, latency and jitter. It is meant to provide the tests to
   run and methodology to verify that a DUT is capable of forwarding
   packets at line rate under non-congested conditions.


2.2 Methodology

   A traffic generator SHOULD be connected to all ports on the DUT. Two
   tests MUST be conducted: a port-pair test [RFC 2544/3918 compliant]
   and also in a full mesh type of DUT test [RFC 2889/3918 compliant]. 

   For all tests, the percentage of traffic per port capacity sent MUST
   be 99.98% at most, with no PPM adjustment to ensure stressing the DUT
 


Avramov & Rapp           Expires April 20, 2015                 [Page 5]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   in worst case conditions. Tests results at a lower rate MAY be
   provided for better understanding of performance increase in terms of
   latency and jitter when the rate is lower than 99.98%. The receiving
   rate of the traffic needs to be captured during this test in % of
   line rate.

   The test MUST provide the latency values for minimum, average and
   maximum, for the exact same iteration of the test. 

   The test MUST provide the jitter values for minimum, average and
   maximum, for the exact same iteration of the test.

   Alternatively when a traffic generator CAN NOT be connected to all
   ports on the DUT, a snake test MUST be used for line rate testing,
   excluding latency and jitter as those became then irrelevant. The
   snake test consists in the following method: -connect the first and
   last port of the DUT to a traffic generator-connect back to back
   sequentially all the ports in between: port 2 to 3, port 4 to 5 etc
   to port n-2 to port n-1; where n is the total number of ports of the
   DUT-configure port 1 and 2 in the same vlan X, port 3 and 4 in the
   same vlan Y, etc. port n-1 and port n in the same vlan ZZZ. This
   snake test provides a capability to test line rate for Layer 2 and
   Layer 3 RFC 2544/3918 in instance where a traffic generator with only
   two ports is available. The latency and jitter are not to be
   considered with this test.



2.3 Reporting Format

   The report MUST include:

   -physical layer calibration information as defined into (Placeholder
   for definitions draft)

   -number of ports used

   -reading for throughput received in percentage of bandwidth, while
   sending 99.98% of port capacity on each port, across packet size from
   64 byte all the way to 9216. As guidance, an increment of 64 byte
   packet size between each iteration being ideal, a 256 byte and 512
   bytes being also often time used, the most common packets sizes order
   for the report is: 64b,128b,256b,512b,1024b,1518b,4096,8000,9216b.

   The pattern for testing can be expressed using RFC 6985 [IMIX Genome:
   Specification of Variable Packet Sizes for Additional Testing]

   -throughput needs to be expressed in % of total transmitted frames
 


Avramov & Rapp           Expires April 20, 2015                 [Page 6]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   -for packet drops, they MUST be expressed in packet count value and
   SHOULD be expressed in % of line rate

   -for latency and jitter, values expressed in unit of time [usually
   microsecond or nanosecond] reading across packet size from 64 bytes
   to 9216 bytes

   -for latency and jitter, provide minimum, average and maximum values.
   if different iterations are done to gather the minimum, average and
   maximum, it SHOULD be specified in the report along with a
   justification on why the information could not have been gathered at
   the same test iteration

   -for jitter, a histogram describing the population of packets
   measured per latency or latency buckets is RECOMMENDED

   -The tests for throughput, latency and jitter MAY be conducted as
   individual independent events, with proper documentation in the
   report but SHOULD be conducted at the same time.




3. Buffering Testing

3.1 Objective

   To measure the size of the buffer of a DUT under
   typical|many|multiple conditions. Buffer architectures between
   multiple DUTs can differ and include egress buffering, shared egress
   buffering switch-on-chip [SoC], ingress buffering or a combination.
   The test methodology covers the buffer measurement regardless of
   buffer architecture used in the DUT.

3.2 Methodology

   A traffic generator MUST be connected to all ports on the DUT.

   The methodology for measuring buffering for a data-center switch is
   based on using known congestion of known fixed packet size along with
   maximum latency value measurements. The maximum latency will increase
   until the first packet drop occurs. At this point, the maximum
   latency value will remain constant. This is the point of inflexion of
   this maximum latency change to a constant value. There MUST be
   multiple ingress ports receiving known amount of frames at a known
   fixed size, destined for the same egress port in order to create a
   known congestion event. The total amount of packets sent from the
   oversubscribed port minus one, multiplied by the packet size
 


Avramov & Rapp           Expires April 20, 2015                 [Page 7]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   represents the maximum port buffer size at the measured inflexion
   point.

   1) Measure the highest buffer efficiency

   First iteration: ingress port 1 sending line rate to egress port 2,
   while port 3 sending a known low amount of over subscription traffic
   (1% recommended) with a packet size of 64 bytes to egress port 2.
   Measure the buffer size value of the number of frames sent from the
   port sending the oversubscribed traffic up to the inflexion point
   multiplied by the frame size.

   Second iteration: ingress port 1 sending line rate to egress port 2,
   while port 3 sending a known low amount of over subscription traffic
   (1% recommended) with same packet size 65 bytes to egress port 2.
   Measure the buffer size value of the number of frames sent from the
   port sending the oversubscribed traffic up to the inflexion point
   multiplied by the frame size.

   Last iteration: ingress port 1 sending line rate to egress port 2,
   while port 3 sending a known low amount of over subscription traffic
   (1% recommended) with same packet size B bytes to egress port 2.
   Measure the buffer size value of the number of frames sent from the
   port sending the oversubscribed traffic up to the inflexion point
   multiplied by the frame size..

   When the B value is found to provide the highest buffer size, this is
   the highest buffer efficiency 

   2) Measure maximum port buffer size

   At fixed packet size B determined in 3.2.1, for a fixed default COS
   value of 0 and for unicast traffic proceed with the following:

   First iteration: ingress port 1 sending line rate to egress port 2,
   while port 3 sending a known low amount of over subscription traffic
   (1% recommended) with same packet size to the egress port 2. Measure
   the buffer size value by multiplying the number of extra frames sent
   by the frame size.

   Second iteration:  ingress port 2 sending line rate to egress port 3,
   while port 4 sending a known low amount of over subscription traffic
   (1% recommended) with same packet size to the egress port 3. Measure
   the buffer size value by multiplying the number of extra frames sent
   by the frame size.

   Last iteration: ingress port N-2 sending line rate traffic to egress
   port N-1, while port N sending a known low amount of over
 


Avramov & Rapp           Expires April 20, 2015                 [Page 8]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   subscription traffic (1% recommended) with same packet size to the
   egress port N Measure the buffer size value by multiplying the number
   of extra frames sent by the frame size.

   This test series MAY be repeated using all different COS values of
   traffic and then using Multicast type of traffic, in order to find if
   there is any COS impact on the buffer size.

   3) Measure maximum port pair buffer sizes

   First iteration: ingress port 1 sending line rate to egress port 2;
   ingress port 3 sending line rate to egress port 4 etc. Ingress port
   N-1 and N will respectively over subscribe at 1% of line rate egress
   port 2 and port 3. Measure the buffer size value by multiplying the
   number of extra frames sent by the frame size for each egress port.

   Second iteration: ingress port 1 sending line rate to egress port 2;
   ingress port 3 sending line rate to egress port 4 etc. Ingress port
   N-1 and N will respectively over subscribe at 1% of line rate egress
   port 4 and port 5. Measure the buffer size value by multiplying the
   number of extra frames sent by the frame size for each egress port.

   Last iteration: ingress port 1 sending line rate to egress port 2;
   ingress port 3 sending line rate to egress port 4 etc. Ingress port
   N-1 and N will respectively over subscribe at 1% of line rate egress
   port N-3 and port N-2. Measure the buffer size value by multiplying
   the number of extra frames sent by the frame size for each egress
   port.

   This test series MAY be repeated using all different COS values of
   traffic and then using Multicast type of traffic.

   4) Measure maximum DUT buffer size with many to one ports

   First iteration: ingress ports 1,2,... N-1 sending each [(1/[N-
   1])*99.98]+[1/[N-1]] % of line rate per port to the N egress port.

   Second iteration: ingress ports 2,... N sending each [(1/[N-
   1])*99.98]+[1/[N-1]] % of line rate per port to the 1 egress port.

   Last iteration: ingress ports N,1,2...N-2 sending each [(1/[N-
   1])*99.98]+[1/[N-1]] % of line rate per port to the N-1 egress port.

   This test series MAY be repeated using all different COS values of
   traffic and then using Multicast type of traffic.

   Unicast traffic and then Multicast traffic SHOULD be used in order to
   determine the proportion of buffer for documented selection of tests.
 


Avramov & Rapp           Expires April 20, 2015                 [Page 9]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   Also the COS value for the packets SHOULD be provided for each test
   iteration as the buffer allocation size MAY differ per COS value. It
   is RECOMMENDED that the ingress and egress ports are varied in a
   random, but documented fashion in multiple tests to measure the
   buffer size for each port of the DUT.


3.3 Reporting format

   The report MUST include:

    - The packet size used for the most efficient buffer used, along
   with COS value

    - The maximum port buffer size for each port

    - The maximum DUT buffer size

    - The packet size used in the test

    - The amount of over subscription if different than 1%

    - The number of ingress and egress ports along with their location
   on the DUT.



4 Microburst Testing

4.1 Objective

   To find the maximum amount of packet bursts a DUT can sustain under
   various configurations. 

4.2 Methodology

   A traffic generator MUST be connected to all ports on the DUT. In
   order to cause congestion, two or more ingress ports MUST bursts
   packets destined for the same egress port. The simplest of the setups
   would be two ingress ports and one egress port (2-to-1). 

   The burst MUST be measure with an intensity of 100%, meaning the
   burst of packets will be sent with a minimum inter-packet gap. The
   amount of packet contained in the burst will be variable and increase
   until there is a non-zero packet loss measured. The aggregate amount
   of packets from all the senders will be used to calculate the maximum
   amount of microburst the DUT can sustain.

 


Avramov & Rapp           Expires April 20, 2015                [Page 10]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   It is RECOMMENDED that the ingress and egress ports are varied in
   multiple tests to measure the maximum microburst capacity.

   The intensity of a microburst MAY be varied in order to obtain the
   microburst capacity at various ingress rates.

   It is RECOMMENDED that all ports on the DUT will be tested
   simultaneously and in various configurations in order to understand
   all the combinations of ingress ports, egress ports and intensities. 

   An example would be:

   First Iteration: N-1 Ingress ports sending to 1 Egress Ports

   Second Iterations: N-2 Ingress ports sending to 2 Egress Ports

   Last Iterations: 2 Ingress ports sending to N-2 Egress Ports

4.3 Reporting Format

   The report MUST include:

    - The maximum value of packets received per ingress port with the
   maximum burst size obtained with zero packet loss

    - The packet size used in the test

    - The number of ingress and egress ports along with their location
   on the DUT


5. Head of Line Blocking

5.1 Objective

   Head-of-line blocking (HOL blocking) is a performance-limiting
   phenomenon that occurs when packets are held-up by the first packet
   ahead waiting to be transmitted to a different output port. This is
   defined in RFC 2889 section 5.5. Congestion Control. This section
   expands on RFC 2889 in the context of Data Center Benchmarking

   The objective of this test is to understand the DUT behavior under
   head of line blocking scenario and measure the packet loss.

5.2 Methodology

   In order to cause congestion, head of line blocking, groups of four
   ports are used. A group has 2 ingress and 2 egress ports. The first
 


Avramov & Rapp           Expires April 20, 2015                [Page 11]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   ingress port MUST have two flows configured each going to a different
   egress port. The second ingress port will congest the second egress
   port by sending line rate. The goal is to measure if there is loss
   for the first egress port which is not not oversubscribed.


   A traffic generator MUST be connected to at least eight ports on the
   DUT and SHOULD be connected using all the DUT ports.

   1) Measure two groups with eight DUT ports

   First iteration: measure the packet loss for two groups with
   consecutive ports

   The first group is composed of: ingress port 1 is sending 50% of
   traffic to egress port 3 and ingress port 1 is sending 50% of traffic
   to egress port 4. Ingress port 2 is sending line rate to egress port
   4. Measure the amount of traffic loss for the traffic from ingress
   port 1 to egress port 3. 

   The second group is composed of: ingress port 5 is sending 50% of
   traffic to egress port 7 and ingress port 5 is sending 50% of traffic
   to egress port 8. Ingress port 6 is sending line rate to egress port
   8. Measure the amount of traffic loss for the traffic from ingress
   port 5 to egress port 7.


   Second iteration: repeat the first iteration by shifting all the
   ports from N to N+1

   the first group is composed of: ingress port 2 is sending 50% of
   traffic to egress port 4 and ingress port 2 is sending 50% of traffic
   to egress port 5. Ingress port 3 is sending line rate to egress port
   5. Measure the amount of traffic loss for the traffic from ingress
   port 2 to egress port 4. 

   the second group is composed of: ingress port 6 is sending 50% of
   traffic to egress port 8 and ingress port 6 is sending 50% of traffic
   to egress port 9. Ingress port 7 is sending line rate to egress port
   9. Measure the amount of traffic loss for the traffic from ingress
   port 6 to egress port 8.

   Last iteration: when the first port of the first group is connected
   on the last DUT port and the last port of the second group is
   connected to the seventh port of the DUT

   Measure the amount of traffic loss for the traffic from ingress port
   N to egress port 2 and from ingress port 4 to egress port 6.
 


Avramov & Rapp           Expires April 20, 2015                [Page 12]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   2) Measure with N/4 groups with N DUT ports

   First iteration: Expand to fully utilize all the DUT ports in
   increments of four. Repeat the methodology of 1) with all the group
   of ports possible to achieve on the device and measure for each port
   group the amount of traffic loss.

   Second iteration: Shift by +1 the start of each consecutive ports of
   groups

   Last iteration: Shift by N-1 the start of each consecutive ports of
   groups and measure the traffic loss for each port group.



5.3 Reporting Format

   For each test the report MUST include:

   - The port configuration including the number and location of ingress
   and egress ports located on the DUT

   - If HOLB was observed

   - Percent of traffic loss

6. Incast Stateful and Stateless Traffic 

6.1 Objective

   The objective of this test is to measure the effect of TCP Goodput
   and latency with a mix of large and small flows. The test is designed
   to simulate a mixed environment of stateful flows that require high
   rates of goodput and stateless flows that require low latency.

6.2 Methodology

   In order to simulate the effects of stateless and stateful traffic on
   the DUT there MUST be multiple ingress ports receiving traffic
   destined for the same egress port. There also MAY be a mix of
   stateful and stateless traffic arriving on a single ingress port. The
   simplest setup would be 2 ingress ports receiving traffic destined to
   the same egress port. 

   One ingress port MUST be maintaining a TCP connection trough the
   ingress port to a receiver connected to an egress port. Traffic in
   the TCP stream MUST be sent at the maximum rate allowed by the
   traffic generator. At the same time the TCP traffic is flowing
 


Avramov & Rapp           Expires April 20, 2015                [Page 13]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   through the DUT the stateless traffic is sent destined to a receiver
   on the same egress port. The stateless traffic MUST be a microburst
   of 100% intensity.

   It is RECOMMENDED that the ingress and egress ports are varied in
   multiple tests to measure the maximum microburst capacity.

   The intensity of a microburst MAY be varied in order to obtain the
   microburst capacity at various ingress rates.

   It is RECOMMENDED that all ports on the DUT be used in the test.

   For example:

   Stateful Traffic port variation:

   During Iterations number of Egress ports MAY vary as well.

   First Iteration: 1 Ingress port receiving stateful TCP traffic and 1
   Ingress port receiving stateless traffic destined to 1 Egress Ports

   Second Iteration: 2 Ingress port receiving stateful TCP traffic and 1
   Ingress port receiving stateless traffic destined to 1 Egress Ports

   Last Iteration: N-2 Ingress port receiving stateful TCP traffic and 1
   Ingress port receiving stateless traffic destined to 1 Egress Ports

   Stateless Traffic port variation:

   During Iterations number of Egress ports MAY vary as well. First
   Iteration: 1 Ingress port receiving stateful TCP traffic and 1
   Ingress port receiving stateless traffic destined to 1 Egress Ports

   Second Iteration: 1 Ingress port receiving stateful TCP traffic and 2
   Ingress port receiving stateless traffic destined to 1 Egress Ports

   Last Iteration: 1 Ingress port receiving stateful TCP traffic and N-2
   Ingress port receiving stateless traffic destined to 1 Egress Ports

6.3 Reporting Format

   The report MUST include the following:

   - Number of ingress and egress ports along with designation of
   stateful or stateless.

   - Stateful goodput

 


Avramov & Rapp           Expires April 20, 2015                [Page 14]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


   - Stateless latency

7.  References













































 


Avramov & Rapp           Expires April 20, 2015                [Page 15]

Internet-Draft    Data Center Benchmarking Methodology  October 17, 2014


7.1.  Normative References

   [1]   Bradner, S. "Benchmarking Terminology for Network
         Interconnection Devices", RFC 1242, July 1991.

   [2]   Bradner, S. and J. McQuaid, "Benchmarking Methodology for
         Network Interconnect Devices", RFC 2544, March 1999.

7.2.  Informative References

   [3]  Mandeville R. and Perser J., "Benchmarking Methodology for LAN
         Switching Devices", RFC 2889, August 2000.

   [4]  Stopp D. and Hickman B., "Methodology for IP Multicast
         Benchmarking", BCP 26, RFC 3918, October 2004.

7.3.  URL References

   [5]  Yanpei Chen, Rean Griffith, Junda Liu, Randy H. Katz, Anthony D.
         Joseph, "Understanding TCP Incast Throughput Collapse in
         Datacenter Networks",
         http://www.eecs.berkeley.edu/~ychen2/professional/TCPIncastWREN2009.pdf".

Authors' Addresses


         Lucien Avramov
         Cisco Systems
         170 West Tasman drive
         San Jose, CA 95134
         United States
         Phone: +1 408 526 7686
         Email: lavramov@cisco.com

         Jacob Rapp
         Hewlett-Packard
         3000 Hanover Street
         Palo Alto, CA
         United States
         Phone: +1 650 857 3367
         Email: jacob.h.rapp@hp.com










Avramov & Rapp           Expires April 20, 2015                [Page 16]