Choose a Distributed Replicated FileSystem

[Alex-31]

Hello,

I need to find et choose a filesystem for a replicated and distributed architecture of files server.
My company is dispatched on more of 10 agency all over the country, and, I would like implement a distributed file system on this kind of architecture.

The goal is to have a Filesystem which appears localy for each users of agency, but, behind, the data have to be replicated between several sites for high availability and fault tolerance.
Each agency is linked with the HQ over a VPN tunnel.

I would like to find a solution on the problem of a network failure... if the internet connexion of the agency breakdown, I would like the Filesystem still be reachable with the local replication.

My principal problem is the bandwith of the agency which can be low so, the replication have to be asynchronous.

For the moment, I'm trying XtreemFS. I have to continu my test because I'm not sure the FS is very stable when a connexion problem occurs.

Any idea ?


Alex

 

[ryanvade]

Ceph may provide what you are looking for: http://ceph.com/
GlusterFS also seems to be a good choice: http://www.gluster.org/
https://www.digitalocean.com/commun...torage-pool-using-glusterfs-on-ubuntu-servers

 

[debian_guy]

I used Ceph in the distributed file system which has over than 100million files at total 10TB of data. Ceph is stable if two of your disks are not broken at the same time. And if two of its broken, it became a nightmare to fix. It caused from the file count i assume. Never experience a network failure problem. Its network should constructed with 10Gigs which i cannot afford at that time.

Their disks must not be sata if it has too many files. Sata disks are too slow for seek time and broken too much. If one of your disk is broken in your system that can be trigger to broke others too, because of its files needs to replicated from other sources and your system will work way more than normal.

It always crashed and searched some ways to fix the problems. Stay away from distributed systems if you are not working them with proffesionally. Make one primary storage and sync it to its backup. That will help you to handle the failures so easily.