[tahoe-dev] some random timing measurements

Brian Warner warner-tahoe at allmydata.com
Fri Jan 25 15:25:35 PST 2008


I'm trying to debug some problems with the newly-implement "offloaded upload
helper" (see ticket #116 for details), and thought I'd record a few notes
about speeds and timings here rather than on a post-it that will get lost on
my desk within a few days.

The scenario is uploading a 20MB file to the helper, which writes it to disk.
Once upload finishes, then the helper encodes and pushes the shares out to
the storage servers.

 hashing the file to compute the encryption key and storage index takes 19ms

  (at 20MB, the file is probably entirely in the kernel's filesystem cache,
  so there's a good chance that the file never actually touches the disk)

 querying storage servers that are in colo takes 17ms each. This includes the
 foolscap round trip time, and the storage server's disk directory lookup.

 querying one storage server that's outside colo (probably on a home user's
 DSL line) takes 234ms

 we currently query storage servers sequentially, rather than in parallel. If
 we didn't have any out-colo servers, the query process would take 204ms


Still investigating the problems.. I'm adding more logging tools to Foolscap
as we go, trying to use the analysis process to improve the tools.

cheers,
 -Brian


More information about the tahoe-dev mailing list