[tahoe-dev] Performance question: reading from/writing to a grid of 4 data centers using 9-out-of-24 coding

[Robin Chen]

Hi Ed and Oleksandr,

Thanks for the prompt replies!

There's no swapping or paging out (si=0 consistently in the vmstat output).
  Plenty of free memory during the reading process.

I'm using 24 VM's (running under OpenStack) in 4 data centers.   Each VM
has two 2 VCPU's, 4GB of memory, and 50GB of disk space (40GB for Tahoe
storage).  All VM's were dedicated to the Tahoe experiments.

I set up a separate Tahoe client to do the performance measurements through
the Thaoe command line tools so that I could collect elapsed time easily.
For example,

$ tahoe create-alias grid
$ time tahoe cp 383MB.mov grid:
$ time tahoe cp grid:383MB.mov t.mov
$ cmp 383MB.mov t.mov

I'll run the same tests through the Web interface and check out the
performance pages - thanks!

Robin

On Mon, Mar 4, 2013 at 11:09 AM, Oleksandr Drach <luckyredhot at gmail.com>wrote:

> *Robin,*
> I think your grid speed is quite good.
> Could you please briefly provide your hardware details?
>
> You can additionally check every download operation on Status Page<http://yournode.com:3456/status/>.
> Click on Status column and you'll find more details<http://yournode.com:3456/status/down-0>
> .
> Timeline <http://yournode.com:3456/status/down-0/timeline> is also may be
> very useful.
>
> You can find further readings on Tahoe-Lafs performance here<https://tahoe-lafs.org/trac/tahoe-lafs/wiki/Performance>and
> there <https://tahoe-lafs.org/trac/tahoe-lafs/wiki/Performance/Old>.
>
> *Zooko,*
> I think adding "real" performance numbers to the Performance page<https://tahoe-lafs.org/trac/tahoe-lafs/wiki/Performance> will
> be really helpful for a lot of people.
>
>
> *Sincerely,
> Oleksandr Drach,*
>
> *e-mail: **luckyredhot at gmail.com* <luckyredhot at gmail.com>*.*
>
>
> 2013/3/4 Ed Kapitein <ed at kapitein.org>
>
>> Hi Robin,
>>
>> I am quiet new too, and i ran into a "out of memory" situation, while
>> reading a large file.
>> Did you check to see if the reading machine starts swapping?
>> ( you can check with top )
>> That could make the machine and the reading real slow.
>>
>> Kind regards,
>> Ed
>>
>> On Mon, 2013-03-04 at 10:45 -0500, Robin Chen wrote:
>> > Hi,
>> >
>> >
>> > I have recently started my experiments with Tahoe and deployed 24
>> > Tahoe storage nodes in 4 data centers (using 9 out-of 24 coding), but
>> > I'm running into a performance problem.
>> >
>> >
>> > I'm new to this email list, so I'm not sure if this problem has been
>> > discussed before or not.
>> >
>> >
>> > I would expect the "reading" performance (reading 9 shares) to be
>> > better than the "writing" performance (writing 24 shares in 4
>> > geographically-distribute data centers), but the reading time is
>> > typically two times longer in most of my tests with files > 300MB.
>> >
>> >
>> > For example, for a file size of 383MB,  I'm getting about 16Mbps in
>> > throughput for writing, but only about 8.8Mbps in reading.
>> >
>> >
>> > Any possible explanation?
>> >
>> >
>> > Thanks!
>> >
>> >
>> > Robin
>> > _______________________________________________
>> > tahoe-dev mailing list
>> > tahoe-dev at tahoe-lafs.org
>> > https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
>>
>>
>> _______________________________________________
>> tahoe-dev mailing list
>> tahoe-dev at tahoe-lafs.org
>> https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
>>
>
>
> _______________________________________________
> tahoe-dev mailing list
> tahoe-dev at tahoe-lafs.org
> https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tahoe-lafs.org/pipermail/tahoe-dev/attachments/20130305/09ee8244/attachment.html>