| 1 |
Your comment on needing to "store *a lot* of data (like 50TB max)" is very |
| 2 |
relative. For sites like ours and many others, *a lot* of data means 100's |
| 3 |
of PB. :) |
| 4 |
|
| 5 |
As to your questions RE: GlusterFS, this announcement has also recently made |
| 6 |
the rounds on the beowulf mailing list though there has not been too much |
| 7 |
conversation about it. Being new, nobody has tried it as of yet, as such |
| 8 |
there are not many comparative studies (if any) out there. |
| 9 |
|
| 10 |
What they claim to be able to do is quite nice. But then again, many vendors |
| 11 |
and other FOSS cluster file systems have made such claims in the past. Only |
| 12 |
time will tell. |
| 13 |
|
| 14 |
The only real thing I can offer is that I will be watching this very closely |
| 15 |
which might lend some credence to Gluster as a potential solution. |
| 16 |
|
| 17 |
Daniel van Ham Colchete said the following on 01/31/2007 12:20 PM: |
| 18 |
> Hello everyone! |
| 19 |
> |
| 20 |
> I've been doing some research to build a gentoo cluster soon. The part |
| 21 |
> I'm giving more attention is the clustered filesystem because I'll have |
| 22 |
> to store *a lot* of data (like 50TB max). For now my choice is the |
| 23 |
> Lustre Filesystem (after they release the 1.6 version). |
| 24 |
> |
| 25 |
> Today I runned into the GlusterFS: The GNU Cluster Filesystem. According |
| 26 |
> with it's website (http://www.gluster.org/) it has: |
| 27 |
> - Per-file replication: you don't need to replicate every OSS like in |
| 28 |
> LustreFS, you can have as many copies of some files as you want. Very good! |
| 29 |
> - Scalability to the Petabytes (it's enough for me) |
| 30 |
> - No single point of failure design. Completely distributed. No |
| 31 |
> centralized meta-data server like Lustre. |
| 32 |
> - It's a userspace filesystem. It doesn't need a specific distribution |
| 33 |
> to run (like Lustre does). You can use Gentoo servers with the lastest |
| 34 |
> kernel version!!!! :-) :-) :-) :-) :-) |
| 35 |
> |
| 36 |
> I would like to know if is there anyone here that tested this *new* |
| 37 |
> filesytem. I know Lustre has other features and probably is faster. But |
| 38 |
> I need a reliable fs for a *large* mail server cluster so I don't think |
| 39 |
> I'll need the other features they have. |
| 40 |
> |
| 41 |
> What do you think??? |
| 42 |
> |
| 43 |
> Best regards, |
| 44 |
> Daniel Colchete |
| 45 |
|
| 46 |
-- |
| 47 |
Brian D. Ropers-Huilman, Director |
| 48 |
Systems Administration and Technical Operations <bropers@×××××××.edu> |
| 49 |
Supercomputing Institute for Digital Simulation and Advanced Computation |
| 50 |
599 Walter Library +1 612-626-5948 (V) |
| 51 |
117 Pleasant Street S.E. +1 612-624-8861 (F) |
| 52 |
University of Minnesota Twin Cities Campus |
| 53 |
Minneapolis, MN 55455-0255 http://www.msi.umn.edu/ |
| 54 |
-- |
| 55 |
gentoo-cluster@g.o mailing list |
Archives