3
votes

I previously asked this ServerFault question: Does anyone have experience with lefthands VSA SAN

The general consensus looks like it does not perform well enough for a production SQL server even at a light load.

So the new question is, How does LeftHand's SAN perform on the HP or Dell dedicated Hardware boxes?

We are looking at the Starter SAN with 2 HP nodes in a 2-way replication, 2 ESX servers hosting a total of 2 Active Directory server, 1 MS SQL server, 1 File Server, and 1 General Purpose Server for things like Virus Scan (All Microsoft Server 2005 or 2008).

The reason I am looking at LeftHand is for the complete software package. I plan to have a DR site and like how the SAN can perform an Async Replication to the offsite location without having to go back to the Vendor for more licenses.
I also like the redundancy built into the Network Raid architecture.

I have looked at other SANS and found different faults with them.

For example, Dell's EqualLogic: Found that although the individual box is very redundant in hardware, the Data once spanned across multiple boxes is not redundant, if a node goes down you have lost the only copy of the data sitting on that hardware (One thing is certain, all hardware fails... When? is the only question.).

I have used an XioTech SAN as well.. Well worth the money BTW, but I think it is overkill for the size of the office I am targeting. The cost to get the hardware redundancy in the XioTech makes it a little out of reach for the budget I am working in.

Thank you,
Keith

3 Answers 3

1
vote

I got several of 'em on HP hardware about a year ago for performance benchmarking testing, and some of the same things I said in the LeftHand VSA SAN question apply here too.

At the time, LeftHand's iSCSI multipathing wasn't truly active/active. Say you have:

  • Four LeftHand servers with 2 1gb network cards each
  • One volume on the SAN for your SQL Server data files
  • One volume on the SAN for your SQL Server log files
  • One SQL Server with four 1gb network cards dedicated to iSCSI

When you run a query on the SQL Server that accesses the data files, you will only get 1gb of read throughput despite the fact that you're using four network cards. The LeftHand devices (and indeed, all of the iSCSI SAN gear I've seen) will only send data from the SAN to one specific MAC address on the SQL Server.

You can work around this by:

  • Using multiple data volumes and log volumes, and manually managing which path is the "primary" path for each back to the server. Even then, you're still only going to get 1gb of throughput for each data file, which restricts your backup performance, DBCC performance, etc.
  • Using 10gb Ethernet.
  • Using network teaming software that actually works. I've tried using a couple of vendors, and I haven't seen them overcome this problem yet.

If you need more than 1gb throughput, then until you hear from someone who's actually pulled it off, and can show it to you (not just say "oh yeah it works great on my l337 b0xx0r") then don't invest your money.

Fiber channel isn't necessarily different: it's just that you can easily get 4gb fiberoptic connections instead of 1gb Ethernet. You still have the same pathing challenges. I'm doing a presentation on this next week at IndyPASS, coincidentally - if you're in the area, swing by.

2
  • Thank you for the informative answer. I have never asked the question to SAN vendors about the return data all going to a single mac on the server. I know that all of the SAN nodes are churning data at the same time, just never thought about the SQL server single NIC being a bottle neck. Do you know how a single 1GB Nic compares to (6) 10K RPM Cheetah in RAID 5 as far as IOs are concerned? May 5, 2009 at 17:43
  • 6 drives in RAID 5 will easily outperform a 1gb nic when doing reads.
    – Brent Ozar
    May 5, 2009 at 19:27
0
votes

Maybe we can get some more input from Brent here, but I don't think the async SAN replication will work with SQL Server Data files, you'll have to do something like database mirroring/ log shipping. Or at least, this is what I've always thought & haven't yet had the privilege to test this myself.

Can anyone confirm or deny?

1
  • Yeah, the SAN replication will work fine, but it'll be a dangerous copy. If replication falls behind, you can have corruption on disk because the data and log files won't be in the proper sync. If you don't do an orderly failover to the other SAN (shut down SQL Server, make sure the remote SAN is caught up, then start up the remote SQL Server) then you can end up with corrupt data.
    – Brent Ozar
    May 5, 2009 at 14:05
0
votes

Well, if the case is that the replication falls behind when being done at SAN block level, you can't just assume that log shipping will not also fall behind. Both replication technologies would be done at some interval, right? Is the concern then lost data vs corrupt data? I wouldn't think the SAN replication would be recognized as usable if the entire differential update hadn't been received. So is the data corrupt or simply lost? If SQL is virtualized, you're not sending just the db or just the trans logs; I am just not sure how the data would be corrupt?

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .