[tahoe-dev] how to see performance numbers (Re: linuxpal updated)

[Greg Troxel]

"Zooko O'Whielacronx" <zooko at zooko.com> writes:

> To see how quickly a server responded to some requests, go to the
> welcome page of a gateway (e.g. http://pubgrid.tahoe-lafs.org/ ) and
> click on "Recent Uploads and Downloads", e.g.
> http://pubgrid.tahoe-lafs.org/status/ . Then choose a recent
> operations and click on it. If that server participated in that
> request then you'll see timings for that server's behavior.

Thanks.  I think the stats could use some spiffing up, but I'm not sure
I get it so I'll put my thoughts in email before ticketing.

I have a main directory aliased to tahoe: on all client machines.  It
contains (recursively) 4 directories and 7 files.  I have been doing
deep check verify/repair/add-lease and losing.  So I kicked off a deep
check and this operation stalled.  I waited for it to finish and got the
following:

  Mutable File Servermap Update Status

      * Started: 13:52:02 05-Oct-2010
      * Finished: 14:00:24 05-Oct-2010
      * Storage Index: [redacted]
      * Helper?: No
      * Progress: 100.0%
      * Status: Finished

  Update Results

      * Timings:
            o Total: 8 minutes
                  + Initial Queries: 54ms
                  + Cumulative Verify: 4.5ms
            o Per-Server Response Times:
                  + [lwlukwtw]: 104ms
                  + [nszizgf5]: 41ms
                  + [pnxevooc]: 94ms
                  + [sp26qyqc]: 76ms
                  + [sroojqcx]: 357ms
                  + [tavrk54e]: 314ms
                  + [varyf6dy]: 190ms
                  + [xflgj7cg]: 142ms
                  + [4rk5oqd7]: 484ms
                  + [5ouv6kib]: 311ms

So one server didn't answer, and the operation timed out after 8 minutes.

Things that would help:

  show servers that are pending in the list, and perhaps put the time so
  far.   at end, leave the servers that didn't respond and make that clearer.

  instead of "Finished: not yet", put "Finished: not yet as of TIME (delta)".

  in 'recent operations', show either total time if complete, or time so
  far in italics or bold, so one can find things that take a long time.

  when done/failed, don't remove the 'sending 12 queries' info -- but
  showing non-responding servers would avoid the need

  scoreboard servers for latency and whether they are working well, and
  show this in the welcome page per server
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 194 bytes
Desc: not available
URL: <http://tahoe-lafs.org/pipermail/tahoe-dev/attachments/20101005/de2a7423/attachment.pgp>