Copyright (c) 1999 CMP Media Inc. Vendors are encouraged to comment on any part of this project. However, Data Comm and the Ircache team reserve the right to change parameters of this test at any time.
Box: Vendor-supplied equipment including cache appliance, computer system, and networking equipment (routers, switches).
Cluster: Vendor equipment plus the testing equipment (Polygraph Unix systems).
Run: A concurrent execution of at least one Polygraph client and one Polygraph server within a cluster.
Experiment: A sequence of Runs with similar configurations and purpose, but with a change of at least one configuration parameter (e.g., ``load'').
Transaction error: Any failure to submit a complete request and/or receive a complete ``valid'' reply (within time limits and other conditions specified by an experiment) as detected by benchmarking software. A valid reply must have all the headers
generated by the server and may include other headers. A valid reply must have exactly the same content as generated by the server.
The "Datacomm-1" workload is similar to the "PolyMix-1" workload for the Polygraph cache testing tool, with modifications and additions described below.
A general description of Polygraph, along with source code and documentation, is available here:
and a description of PolyMix-1 is available here:
The Datacomm-1 workload differs from PolyMix#1 in these ways:
* HTTP persistent connections will be used. The number of requests handled per client TCP connection is to be determined by a Zipf distribution, with an upper limit of 64. For the server side, it is a Zipf distribution with a limit of 16.
* Runs will be four hours in duration instead of one hour.
* An "object lifecycle model" will mimic PolyMix#1 behavior. This model determines the values for Date, Last-Modified, and Expires headers.
Because of the increased run duration, there will be fewer overall runs. Each box will be tested at TWO different request rates. The request rates to be tested will be supplied by the vendor.
To execute this workload, Polygraph version 1.2.1 or later will be used, with the following command-lines:
--verb_lvl 4 \
--ports 1024:30000 \
--proxy $Proxy \
--launch_win 1min \
--rep_cachable 80p \
--pconn_use_lmt zipf:64 \
--nagle off \
--robots 1 \
--req_rate $RR/sec \
--dhr 55p \
--pop_model unif \
--tmp_loc none \
--cool_phase 1min \
--port 80 \
--verb_lvl 4 \
--idle_tout 135sec \
--pconn_use_lmt zipf:16 \
--nagle off \
--xact_think norm:3s,1.5s \
--obj_with_lmt 100p \
--obj_life_cycle const:2year \
--obj_bday const:-1year \
--obj_expire 100p=lmt+const:1 \
1) No-Proxy: a no-proxy test will be executed to verify the Cluster equipment. The no-proxy test will be one hour in duration. Results of the no-proxy test will not be included in the article.
2) Filling-the-cache: The cache must be in a full state before the DataComm-1 experiments may be executed. A cache is "full" when its disk utilization stops increasing. Results of filling-the-cache will not be included in the article.
3) Datacomm-1: TWO runs of the Datacomm-1 workload, at two different request rates. Failed runs may be repeated, if time allows.
The following metrics may be reported in the Data Comm article:
* request rates specified by the vendor
* success or failure of a run. If a run initially fails, and later succeeds (at the same request rate), the failed run is not reported.
* response time mean and median values, averaged at no less than one-hour intervals.
* document-hit-ratio and byte-hit-ratio values, averaged at no less than one-hour intervals.
* Percentage of transaction errors.
* Once testing has begun (i.e., the no-proxy test), the vendor is not allowed to make any changes in the Box configuration or software versions.
* In the event of hardware failures, vendors will be allowed to swap in new equipment and continue testing, strictly on a time-permitting basis. In the case of a hardware failure where time constraints prevent retesting, this will be noted in the article. Vendors are strongly encouraged to bring spare interfaces, chassis, disks, etc., as a contingency for any hardware failure.
*Participants will receive all test results and have a chance to comment on their results before publication.
*Participants are not permitted to withdraw or suppress results once testing begins.