Been reading (here) and other about GLUSTERFS.
Anyone have first hand experience going with DFSR vs. GLUSTERFS or switching between the two?
DFSR = Microsoft's Distrubted File System Replication or DFS on top of their file system replication technology.
The application is not web based. It's traditional file server role; clients are all Windows7 64bit currently accessing an Linux + Samba3 + OpenLDAP system in a DOMAIN role so machines and workstations are authenticated...
I have budget to get two new 2970s with following specs: RAM: 8GIG, maybe 16GIG, 6 Disks (2 in RAID1 for OS, 3 in RAID5 for files, & 1 hot spare for either raid volume).
Files are mostly office type files (Word, Excel, JPG, PDF). Approximately 30 clients attached to current Linux file server today. Roughly a million files or so; they are attempting to go paperless for a portion of their operations and moving all company prints (they make things from prints) to an online retrieval type setup (in case this matters -- point being they have many small files). On a somewhat realted note, we notcied that SMBD processes are consuming higher memory presumable due to the number of files and directories under a common share. One of the other tasks we're doing is scattering the document files into a more tree-like layout under the share to reduce the memory load.
Presently we rsync between file server and a stand-by server but fail-over is manual.
Goal is to get a more automated fail-over/fail-back scenario if possible without much more expense than the hardware we are budget to purchase.
both servers will be in same physical IT room attached to same switch. Switch is a GigE Cisco.
I'm interested in hearing any real world feedback on your use or testing with GLUSTERFS in a similar setup or DFSR. And would be exstatic if you have feedback going between these two technologies and what prompted the switch.
=============================
as a follow up:
We wound up going the Microsoft solution with DFS replication in a full-mesh configuration and I just have to say it works very well. I had my doubts; I've long hated Microsoft but this work very well. The thing that sealed the deal was an article written by a non-Microsoft person who spent considerable time in researching a similar setup and found with Win2K8 R2, specifically, major enhancements were done to speed up DFSR and posted a number of registry hacks to take advantage of the latest DFSR performance tweeks. Out of the box, everything worked rather well. The only pitfall I got myself into was copying all the old data over to the new DFSR shares without first staging it. I got greedy. There were a number of shares to be copied ranging from a few 100 MB to several GB. I did a couple 15GB copies no problem but one that was close to 300GB just tanked and eventually timed out & server became unresponsive over the network. I had to kill the copy and just copy to a non replicated directory and then "copy" not "move" the finished result into the replicated share; deleting the staged data when I was done.
I may look at GLUSTERFS when budget doesn't permit but if anyone wants any details, hit me up.
I have built a simple Samba server using GlusterFS as the storage backend and doing so was extremely easy. I followed the documentation on the GlusterFS website, mounted my GlusterFS volumes on the Samba server and was able to share them quite easily. There were no performance issues with the combination, and GlusterFS is capable of handling 6750 exobytes so you won't have any storage limits to worry about. GlusterFS also supports distribution, replication and striping, as well as distributed replicated and distributed striped. I do not have any experience with DFSR, but by default I steer away from Microsoft anything.
I am new too in glusterfs and use it as postgresql active/passive server with heartbeat. I am in testing level right now and it seems very successful. and for future prospect I want to use it with multiple load balanced static java servers. I hope that will not disappoint me.