What would you recommend for a medium office fileserver? By medium office I mean roughly 100-150 users that connect to the fileserver sporadically during the day. We're talking about 2TB of data and rising. File sizes ranging from a lot of small files (doc, xls, pdf, fonts) to large files (HD videos and huge PSD). Users connect to these servers using both Macs and PCs.
We're currently using a few separated servers of a few GB each but we'd like to centralize the management of that, leave some place to growth and simplify the location of the files to the users.
At equal pricing, what would you suggest? We could get something like a Dell (R410 or NX300) with RAID5 or the HP equivalent and we would get good pieces, support and roughly 4TB of usable space. Or, for the same amount of money, we could put together a few (4-5) home-made rack-mounted servers (Supermicro board, older Intel processors, 2 or 4 GB or RAM and some basic Caviar Black) and cluster everything through GlusterFS or similar. This would remove the enterprise support option and would move most of the work in our hands but would give us far more usable disk space for the same price.
Both possibilities have their own pros and cons. I'm not really looking for arguments for/against enterprise support, I'm more looking for this:
- Any experience with home-made clusters?
- Any arguments on the risk of using commodity hardware?
- Any points on the performance of such system?
- Any recommendation on alternate solutions that gives place for growth?
Go for an "ISS" (Industry Standard Server) from HP, Dell or IBM. Use RAID5 or 6. Pick a server with plenty of drive slots, or, if your budget allows, get an external SAS drawer. Keep the OS on separate disks (mirrored), so that future disk expansions don't involve a rebuild. 10K disks are fine.
Don't worry about CPU - go for the lowest spec possible (file servers only wake up during an anti-virus full scan or file backup). Memory isn't too much of a concern either. 2GB will be plenty (not sure you can get a server with less now).
Your biggest concerns will be file management (quotas, single instancing, etc). If going down the Windows Server route, look at File Server Resource Manager in Win2k8 R2.
Any experience with home-made clusters?
Some with OpenFiler around the same user load that you describe, but I haven't clustered it together. Didn't need to. You can run the kind of load you've described quite comfortably with a single well-specced SuperMicro-chassis build running OpenFiler or FreeNAS. As long as you ensure that you stack it with 8Gbs+ RAM (for fileIO caching), a competent controller with 1Gb Flash or Battery-Backed Write Cache, and sufficient spindles to serve the IO load with acceptable seek times, then you can shave a good chunk of cash from the big brands.
I wouldn't focus on clustering so much for this level of usage. Just because you could purchase multiple identical SuperMicro builds for the same price as a single HP, doesn't mean you'd gain anything by doing so. However if you feel it's warranted, then DRDB is worth a look.
Any arguments on the risk of using commodity hardware?
The biggest problem I've found is getting a suitable monitoring solution to ensure you're notified if any components in the system fail. HP have Insight Agents, Dell have their OpenManage etc. This is really important because you need to know the second a hard drive fails. Also beware the peril of using non-enterprise drives for enterprise RAID sets, and the time-out problems that can occur due to a lack of TLER on the drives.
Any points on the performance of such system?
There's really nothing special in a branded system that puts it's performance above an unbranded one. The key is that they're usually specced by experts for particular workloads and ship with components that are guaranteed to work well together. Good quality array controllers and NICs with offload capabilities are advisable.
Any recommendation on alternate solutions that gives place for growth?
As above FreeNAS/OpenFiler are worth a peek. DFS on Windows 2008 R2 is also worth a look as it's been completely overhauled from previous versions and appears to be a much stronger solution than before. No reason you couldn't run 2008 R2 on a custom server build, if your components have appropriate drivers.
How about 2 Windows file servers and DFS?
Small workstation level computer, 16gb ram, good raid card, SAS backplanes and 2.5" discs - I can get about 30 discs into one small tower ;)
DFS, using replication.