In case the SRM server is busy, how to estimate how long it will take to fulfill the request?

Some comments by Arie Shoshani and Alex Sim.

The issue of of SRM_busy means is related to request estimation. We concluded that any descriptive measure of how busy an SRM is (such not-busy, slightly_busy, very_busy, etc.) is not good enough, since these will have to be quantified with performance measures. So, instead, we think that what makes sense is to return time estimations.

Now, there are two options:

  • 1) return time estimations when a request is made,
  • 2) return time estimates when an srmPing is done.

For option 1, we think the time estimates should be: request_begin_time_estimate, and request_end_time_estimate.

For example, if the request had multiple files, request_begin_time_estimate is the time when the first file will be made available, and the request_end_time_estimate will be an estimate when the last file will be made available. This time estimates will take into account whether the requested files are already in cache, as well as other requests in the queue, and service policies (such as round robin service for requests, etc.). Another example is a request for a single file. In this case the begin_time and end_time will be the same.

For option 2, we think the time estimates should be: request_begin_time_estimate, and end_time_estimate_to_complete_all_current_requests

In this case, a request is not provided to the srmPing, so the SRM can return an estimate of how long it will take for the first file if a request was issued (assuming that the file is in cache) - request_begin_time_estimate. The end_time_estimate_to_complete_all_current_requests is an estimate of how long it will take to service all pending requests - it is a measure of how busy the SRM is.

Note that taking this approach means that all the properties of the SRM (cache size, memory size of machine, transfer speed, policies, etc.) are not explicitly provided. The time estimate account for all these based on previously collected statistics by the SRM, or whatever method the SRM wishes to make an estimate. This will be left for each SRM implementation to choose.

-- ElisaLanciotti - 12 Feb 2009

Edit | Attach | Watch | Print version | History: r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r1 - 2009-02-12 - ElisaLanciotti
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    LCG All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback