Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

here is the scenario - there is a small set (200gb around) of data that I HAVE to keep available. Those are basically shared VHD images that serve as master images for a lot of our VM's - they then run in differential discs off those. The whole set is "mostly read only". In more detail: A file that IS there and IS used will NEVER change. I may delete files (when absolutely not in use) and add new files, but a file that is there once gets read protection set and that it is until it is retired.

Obviously, I need as much uptime as possible. SO FAR we run that by having this directory local on every Hyper-V server. Now I think moving this into our storage fabric. Due to the "it HAS to be there" I pretty much want a share nothing architecture.

DFS would be perfect for this - a file never changes, so replication would work nicely. Folders could be replicated to a number of servers, all would reference them from there. Now, that hyper-V supports SMB that could be a good idea to isolate these on a number of servers - we try to move into a scenario where the storage is more centralized.

Server 2012 supports always on shares, but it seems that this only works with a clustered disc behind. Is there any way around this for read only file stores? All documentation points to stuff like a shared JBOD - but that would leave me open for file system corruption. I really plan to go quite separately here, vertically - 2 servers, both with SSD only for this, both with their own 2000W separate USV, both with enough bandwidth to handle everything thrown at them (note to everyone tinking this is 10G - this would be SLOW and EXPENSIVE compared to a nice Infiniband backbone). The real crux is that this is an edge case obviously - as the files are read only once in use.

share|improve this question
I need some clarification; why doesn't DFS-R fit here?IIt's not zero second transparency, but it works good.. – pauska Oct 6 '12 at 19:55
Because it means that when Server A Failes, all VM hosts that reference master images and are served from Server A will suspend... or? Read error = stop of VM - perfect disaster. – TomTom Oct 7 '12 at 5:08
Ah, I didn't get the part about these being active VHD images. – pauska Oct 7 '12 at 11:52
Yeah. The alternative is stuffing a 120gb SSD into every Server and use that - with DFS-R to handle the autoamtic Distribution of updats - to reference the master Images locally. – TomTom Oct 7 '12 at 12:17
up vote 0 down vote accepted

I don't think this is doable. You can't have failover transparency without using shared disks. How would the other node know which files the first node was serving without using locks?

The only other option I can think of from the top of my head is to use Fault Tolerance inside vSphere, but that has very high requirements in terms of processing power, network bandwith and latency. I've never encountered anyone doing fault tolerance on less than dedicated 10gig pipes.

share|improve this answer
Let me ask you another question - why would the first Server Need any locks when THE WHOLE FILE AND SHARE IS READ ONLY? ;) Those are master iamges, not actively written once put into production. We occasionally delete files - once not referenced anymore - and occasionally add new master Images, but a file, once written, is never changed. So lock tracking is simply not needed at all, you know. – TomTom Oct 7 '12 at 12:18
I'm not saying that your idea is flawed, i'm just telling you how clustered file servers work. – pauska Oct 7 '12 at 13:43

Have you thought about something like a small NetApp filer - they have two 'heads', one take over from the other in a failure scenario and within a short enough time to cause no disconnections. We use them extensively and have never had an outage due to problems with the filer. They're not free though ;)

share|improve this answer
I have htought about many things, but These machiens i talk in question serve only VHD master Images and serve as lasat resport / anchor Servers and must be independant for this Scenario (DC, Virtual Servers for other DC's). I am putting up a quite large storage aray for the "main production" site, though netapp was way unreasonably in the disc range I needed (we talk right now of a Setup for 112 discs with 2x10gb nics that form the main storage). This question is very particular only for the "corner" machines. – TomTom Oct 7 '12 at 14:47

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.