Re: Question about distributed FS for high-performance I/O [debian-hpc]
On 29-01-2016 00:59, Dmitry Smirnov wrote:
On Thu, 28 Jan 2016 03:46:59 PM Fabricio Cannini wrote:
I've setup a glusterfs cluster to do what you want without much fuss.
https://packages.debian.org/jessie/glusterfs-server
https://packages.debian.org/jessie-backports/glusterfs-server
How well did it work for you?
Like I said, it was a long time ago. Long as in mid-2010.
Surely things must have improved since then.
(at least I hope so)
I spent like half a day reading and fiddling with it to get a working
setup, without "wizards" or anything like that, IMO, Not that bad for a
complete newbie. I ended up with 20 nodes running gluster. Performance
was ok.
IIRC gluster has a few replication modes: striping (like raid), whole
file copy ...
Also, it does have a metadata server, and you can setup how many and
what nodes will be metadata servers.
I'm writing this from memory, please check the documentation
I know GlusterFS does not care for data
integrity and I'm concerned about potential inconsistencies due to lack of
metadata server...
Nowadays an alternative would be ceph, but I've no experience with it.
No no no. I strognly advise against Ceph. It is very sophisticated and
fragile with huge and messy code base... Ceph is very unreliable, very slow
and difficult to set up. Ceph couldn't care less for data integrity and in
loing term data corruption is inevitable. I'm talking from experience. Please
stay away from Ceph, it does not worth the effort.
Like I said, I've no experience with ceph. ;)
For curiosity's sake, when/what version of ceph did you used?
[ ]'s
Reply to: