What is this?

We're working on the changes needed in policy, technical configuration, up to the storage model in the data storage system foreseen for LHC. That is, how to get the most (or enough) efficiency out of the current equipment; what parts of the tape/disk/client/networking architecture do or will need upgrading in the following years so that LHC data requirements can be met; and what bigger or smaller changes need to be introduced into the model so as to get better efficiency.

In short: Current and future hardware can do much more than what we use it for (eg, low efficiency). We could choose to live with this efficiency rate, buy more tapes and the current model would serve. However, by modifying this model we hope to gain in efficiency and therefore lower the cost of meeting the needs of LHC.

Who is involved?

As a development project, it is handled by ADC group with close collaboration from FIO, and CS and the experiments. Subscribe the mailing list, attend the meetings and get a Wiki account if you want to participate.

If you edit this Wiki to add your suggestions (on a separate page) or comment on / modify others please make sure it is linked below...

What is the current status of events?

• First meeting was done and quite successful, with already a lot of important information exchanged
• This wiki page and the mailing list have been started (Archives)
• "back-of-an-envelope" calculations were called for for the various enchancements proposed
• also, a synthetic test procedure was called for to help the evaluation of these enhancements

NEXT MEETING: n/a

Proposed "building blocks" or enhancements

• FileMarks
• LocalCacheForTapeServers
• DiskServersSingleDirection
• EnhanceNetworking
• BuyBlackBox
• ArchiveFiles

Proposed complete solutions to be tested

• LargeTapeServer (involving FileMarks, LocalCacheForTapeServers and EnhanceNetworking, servers ~5 tape drives)
• MidTapeServer (involving FileMarks, LocalCacheForTapeServers and EnhanceNetworking, servers 1-2 tape drives)

Tests in progress

• TestContinuousStream - to avoid file marks (at least, to see if we can do that) until 16 May 2005
• DiskServerIONice - I/O prioritization on disk servers (using 2.6 kernel)

-- AndrasHorvath - 25 Apr 2005