[Gluster-devel] Mail cluster storage benchmark

[Basavanagowda Kanur]

Daniel,
  Replies inline.

On Tue, Nov 11, 2008 at 12:10 AM, Daniel van Ham Colchete <
daniel.colchete at gmail.com> wrote:

> Hi yall!
>
> for the last few days I have been developing a software to benchmark a mail
> server cluster. The reason for this is that first I couldn't find one
> online, second using sequential tests on any filesystem won't give you
> what's best for you because with e-mail everything is happening in parallel
> and you won't see if one option wins on parallelism (tar, rsync, cp, find,
> ls -la, etc all are sequential tests), and third is that there is no better
> way to produce a similar read/write IO pattern. This software will be
> released GPL next month.
>
> Thursday I'll be arriving at my data center (at another continent by the
> way), I'll be there for 13 days and my main objective for these days is to
> put GlusterFS up running there. The current setup is not good enought and it
> doesn't scale anymore. Other things are secondary, although there are a few.
> Throughout those 13 days I will be using my software to study many many
> options for the storage servers, for example:
>
> Filesystem options: Ext3, XFS, ZFS (OpenSolaris)
> RAID options: RAID 10, RAID 5, RAID 6, RAID 0 + (ZFS:RAID-1, ZFS:RAID-Z,
> ZFS:RAID-Z2)
> GlusterFS: 1.3/1.4, Unify/DHT, BDB/Posix, IO-Cache, Read-Ahead,
> Write-Behind
> Network file systems: NFS/GlusterFS (splitting directories with NFS is a
> escape plan).
>
> Besides other options involving e-mail hosting, like what RDBMS I should
> use to do user authentication, etc...
>
> I have 4 nice mail servers (storage clients) and two nice storage servers
> (4TB each) available for the tests. My final setup is exactly the double (8
> mail servers, 4 storage servers). Right now I have about 9k users checking
> their e-mails every 5 minutes, running with HA-NFS, but this setup is my
> current bottle-neck and it doesn't scale (I can't just add more storage
> servers). I expect to put 15k users on the new setup but I wish it could
> grow bigger.
>
> Of couse I'll send all the results to the list, put on the wiki too (NFS x
> Best GlusterFS for mail storage, how is this going to end?). I would like to
> ask you devs and users a few questions:
>
> (Devs and users) How do I get started with OpenSolaris and ZFS? What should
> I be looking for with my bechmarks? What do I have to study on ZFS
> performance optimization? I have no experience at all here.
>
> (Devs) Can you send me an example on the DHT? The DHT is what I have always
> dreamt about, I'm really betting on it.


volume dht
    type cluster/dht
    subvolume <sub-1> <sub-2> <sub-3>
end-volume

please note that you need to have support for extended attributes on your
storage.


>
>
> (Devs) I've seen by the changelog that the 1.4 tree development is going
> really fast. Bug are being fixed everyday and this is great! I hope I'll
> help to find a fix a few too in this process. So, say 1.4 is much much
> better for me (I'm betting on that), when do you guys expect an stable
> version? How easy is to migrate from 1.3-Unify+AFR+Posix to 1.4-DHT+AFR+BDB?


It is pretty straightforward. :)

DHT servers the same purpose as unify, distributing files over subvolumes.

You need to have Berkeley DB library for BDB to work.


>
>
> Thank you all for all the help!
>
> Best regards,
> Daniel van Ham Colchete
> (IRC: vanham)
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at nongnu.org
> http://lists.nongnu.org/mailman/listinfo/gluster-devel
>
>


-- 
hard work often pays off after time, but laziness always pays off now
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-devel/attachments/20081111/2251d975/attachment-0003.html>