This page contains notes of the organizational meeting that took place on August 8, 1999 in Boulder, Colorado. Old preparation materials for the meeting are available elsewhere. Preparation materials give detailed overview of many issues discussed at the meeting.

The following companies sent their representative to the meeting:

CacheFlow (Nakul Saraiya) Cisco (Stew Forster) Compaq (Jeff Potter) Dell (James Jech) Entera (Jeffrey Wheelhouse) Eolian (Pamela Maloney) Falcon Systems (Pei Cao)

Hewlett Packard (Jay Veazey) InfoLibria (Abdelsalam 'Solom' Heddaya) Lucent Technologies (Leland Beaumont) Microsoft (Jose Sua) Network Appliance (Karl Swartz) Novell (Ron Lee)

IBM and Netstor intend to come to the cache-off but could not attend the meeting.

1. cache-off logistics (time and place)
2. test suite (what kind of tests will be run)
3. rules
4. budget (participation fees and such)
5. magazines participation and press coverage
6. miscellaneous

Here we summarize important decisions made during the meeting. Virtually all decisions were adopted using majority vote. All 13 vendor representatives had the right to vote while Polyteam members did not vote.

• Cache-off date was set to January 17th, 2000. See deadlines page for the comprehensive list of deadlines.

• There will be no bail-out option for the second cache-off. By starting the first official cache-off test, a participant promises to complete the required test sequence and allows publication of all the results whatever they are. As usual, participants will have a chance to run unofficial experiments before starting an official sequence. Official tests are described elsewhere and include filling-the-cache, downtime, and performance tests.

• Vendors found it infeasible to reliably verify that a caching solution is a full-featured cache (i.e., can handle more than just Polygraph workloads). No such checks will be performed during the cache-off.

• Polyteam will not attempt to measure and report ``maturity'' of products under test.

• A product under test must be available for purchase within 3 month of the cache-off start date.

• When results are published in a trade journal, Polyteam will not select or assist in selection of any kinds of ``winners''. Vendors agreed with Polyteam that such selections are usually meaningless. Magazines are free to make such selections if they have to.

• Each cache-off participant will be charged a participation fee. The fee will be based on number of PCs needed to simulate the required request rates. A non-refundable deposit of $1-2K must be submitted to Polyteam.

• Given sufficient resources (space, power, cooling, time, etc.), there will be no limits on the number of participants. If there is a shortage of resources, Polyteam will try to allocate testing slots in a fair manner, such that all unique ``solution providers'' have at least one slot.

• Polyteam will attempt to use dummynet tool to simulate packet delays, packet loss, and bandwidth limitations where applicable:

  Delays		Packet Loss		Bandwidth
  Client Server yes no no no		Client Server yes yes no yes		Client Server 33/56Kbps unlimited unlimited unlimited

  Vendors acknowledged that using dummynet may require FreeBSD operating system to accurately reproduce cache-off results. However, since network traffic shaping is very important (and since there is no portable tool with the desired functionality) using dummynet was approved.

• If the Object Life Cycle model will be included in PolyMix-2, there will be about 10% of objects with Expire header fields, 70% of objects with Last-Modified-Time headers, and 15% of IMS requests. There will be also some ``reload'' requests (reloads will be independent from object staleness). The values quoted above may be adjusted and more headers/request types may be added.

• Performance tests must report at least 25% document hit ratio for all reported test phases (averaged over a phase).

• The proxies must set MSL value to at least 30 seconds, yielding 60 second 2MSL values. Polyteam may test MSL settings on proxies.

Please note that many more issues were discussed.

Here are a few pending items/issues.

• Vendors are to send Polyteam preliminary number and request rate of products they intend to bring to the cache-off (assuming no artificial participation limits). Based on that information, Polyteam is to do make ballpark estimation of the participation fee and other logistic details.

• Vendors are encouraged to contribute to finding the ``right'' parameters to the Object Life Cycle model and improving the model overall. If suitable model is not found until September 13th, PolyMix-2 will not have an OLC component and related modules such as IMS requests.

If you have attended the meeting and feel that our summary is incorrect or incomplete, please let us know.