Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I've currently got 3 ESXi servers (free version), using direct attached storage. I want to move to shared storage, because direct storage is very limiting. There are about a dozen VMs, most are fairly low requirements (Exchange is the only one with much IO).

Unfortunately, budget constraints mean I can't get a proper SAN. Can't even get something reasonable off ebay.

Basically, I have two choices. One is a Netgear ReadyNAS 3200. I'm kind of against this, because it's only got 2 gigabit ports, and if you're using one of them for management, that leaves 1 port for all our VMs. Not good.

The other is to pick up a Dell server (such a a T410), fairly low spec, and then to get a quad port gigabit card, and put openfiler on this. For the same kind of price, I can get a dual core xeon, 4GB of RAM, and 6TB of storage (which is enough) on a RAID card, with dual PSUs.

Has anyone done something like this before? Can anyone comment on the performance I might get for this kind of setup?

And yes, I know you should have a proper (new) SAN, dedicated network, replication, etc, but that simply isn't an option. Thoughts?

share|improve this question
1  
I think you should look at the readynas again, you can bond the ports to aggregate traffic and still use them for the management interface. –  ErnieTheGeek Jun 6 '11 at 18:41
    
What's limiting in regards to direct storage, especially since you're using the free version of ESX, which can't take advantage of the benefits of shared storage (vMotion, HA, etc.). –  joeqwerty Jun 6 '11 at 19:12
1  
joeqwerty's questions is key I think. If you aren't clustering anyway, I'd personally rather have direct local storage than either of the other 2 options you've mentioned. –  icky3000 Jun 6 '11 at 19:17
    
You can do HA/vMotion etc if you have vCenter. But I guess the OP doesn't have this? –  George Hewitt Jun 6 '11 at 20:19
    
@Ernie - I've read a lot of people saying it's best to keep iSCSI separate, so was hoping to put the SAN on a dedicated switch, isolated from the main network. If both ports are on that, I can't easily manage it. –  Dentrasi Jun 6 '11 at 20:54
show 2 more comments

4 Answers 4

It's quite difficult to answer your question without knowing almost anything about your workload; it will work, sure... but how it will work in your specific scenario depends entirely on how you are going to use it; you should definitely run some performance monitoring on your current storage and then try to find some benchmarks for the two solutions you're evaluating.

I'd personally go for the Dell server, you'll get much more flexibility and (probably) performance; just try to get one with a lot of disk bays... and avoid at all cost creating a single big RAID 5 array, as it can and will severely damage performance if you put many VMs in it; RAID 10 is much faster, and having two or more arrays is a real improvement.

Also, a side note: if you want to be able to actually use that shared storage to move your VMs between the various servers, you'll need a Virtual Center. And then, ESXi will cease to be a free product. Without a VC and proper licensing, you can't actually do much more than now by simply replacing local storage with a shared one.

share|improve this answer
add comment

The T410 you are describing is probably the same price as getting a entry level iSCSI from Promise, such as the M210p

Also, you would be surprised as to how far one iSCSI port can go. I've seen almost 10 VM's running off of an iSCSI with one port and even when pulling backups I wouldn't reach more than 50% of the capacity.

On the other hand, I also built an iSCSI server using Linux and found it pretty straight forward.

To me the advantage of a iSCSI SAN is the hardware RAID, if you want to achieve the same level of quality on the Dell server, be ready to spend a couple of hundreds of bucks on the HW RAID card.

I also agree with Massimo on the RAID level, 10 is the way to go for VM's IMO.

share|improve this answer
add comment

You might want to check out Windows Storage Server or OpenFiler. Obviously, neither are ideal, but on a budget you have to do what you have to do.

share|improve this answer
2  
WSS is only available to OEMs. You can dl it via technet, but you can't use it in production legally. Also, it kinda sucks compared to the Linux flavours (openfiler/freenas) –  Mark Henderson Jun 6 '11 at 21:25
    
Agreed with @Mark on this. I'd highly suggest not going with WSS. Not too long ago, we were stuck in a catch-22 WRT support on WSS. Microsoft wouldn't support us (even if we paid them), cause it was supposed to be an OEM-supported product, and HP plain didn't have (or wasn't willing to provide) the software troubleshooting chops to debug or resolve what we were seeing. We ended up going "unsupported" and reinstalling vanilla Windows Server on that system so we could get decent support if needed. –  EEAA Jun 7 '11 at 13:24
add comment

I agree with Massimo that an answer to this question depends heavily on the workload. I have personally used Openfiler in production in a low load fileserver. Take in mind that Openfiler doesn't support SCSI reservations and this can cause some problems. Check out this KB for more info.

iSCSI can be a cheap SAN solution, just make sure that the underlying harddrives and controllers have a decent performance. Two bonded 1 GB NICs can give you about 200 MB/s of throughput, but again, this depends on the type of I/O (random, sequential) of the VMs.

share|improve this answer
add comment

Your Answer

 
discard

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.