[tahoe-dev] call for volunteers for volunteergrid2

Greg Troxel gdt at ir.bbn.com
Fri Jan 14 13:50:19 UTC 2011


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

> On Wed, Jan 12, 2011 at 6:56 AM, Greg Troxel <gdt at ir.bbn.com> wrote:
>>
>> "Zooko O'Whielacronx" <zooko at zooko.com> writes:
>>
>>>> [grid with servers coming and going
>>
>>> I would be happy to learn the results of this experiment.
>>> I've filed a few tickets that will need to be fixed first.
>
> Those are good tickets, and would certainly help with updating mutable
> objects, but I don't necessarily agree that they are prerequisites.
> Pandelis's ad hoc grid might work even without these tickets, if the
> grid operators figure out how to make it work. Or maybe not -- I would
> like to find out.

For others who weren't in the ticket discussion, I am referring to:

  http://tahoe-lafs.org/trac/tahoe-lafs/ticket/1209
  http://tahoe-lafs.org/trac/tahoe-lafs/ticket/1210

My experience in the pubgrid is that I try a deep-check --repair
--add-lease more or less daily.  Almost always objects need repair,
typically because a server on which yesterday's repair placed a share is
not there.  Then, the next day it might be back, and a repair is
triggered again because of a stale share of seqN-1.   (This is the good
case; sometimes there are too few servers taking shares.)

I just did a repair and of 8 directories, 7 were ok and 1 was
regenerated because my last repair didn't contact server "trelpancake"
and seq32 was there, instead of seq33 everywhere else (of which all 10
shares were present).

Does this happen to you?

I guess basically I meant that I expect this kind of trouble and more
From servers with low availability (even 80% with poisson on time of 4
days and poisson off time of 1 day causes daily churn with 10 servers).
It seems easy to address, and fixing it will expose the next layer of
complexity in high-churn environments.

>> FWIW, so far I'm heading for a private (invitation-only to people I
>> already know) grid with a constitution like volunteergrid2.
>
> Very interesting! Could you please summarize for tahoe-dev what you
> learn about operational issues?

Certainly, but I don't expect anything useful for at least several
months.  I did order a 2T disk and a soekris box, though.  I already
posted my thoughts about the issues I expect to encounter (I know you've
seen this already; it hasn't been updated):

  http://www.lexort.com/blog/tahoe-lafs.html
-------------- 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/20110114/72d7a5b7/attachment.asc>


More information about the tahoe-dev mailing list