Faithful reader(s), it’s been a while! I’ve been busy preparing for some big transitions. I’m also getting better at MDX queries, tweaking SSAS-based reports to more accurately reflect the business rules. But enough about that, on with the post!
In which we doubt the SAN
A storage area network (SAN) is a management & administration solution, not a performance solution.
-someone wiser than me
SANs are wonderful technology. They inspire all kinds of geekery and are purported to solve all your storage woes. But there’s a catch: they’re expensive. Not just as a capital expense, but in maintenance and licensing costs. And if you ever want to upgrade it, like add some more drives to a particular tier/pool — fuhgeddaboudit.
So what do we do with SQL on a SAN? Well, it has tiers, right? Slower storage with huge capacity, faster storage with less, etc. We put the data files (heavy random read workload, typically) on the pool optimized for that kind of I/O pattern. We put the TLog files (heavy sequential write workload) on the pool best suited for that. And what about good ol’ TempDB? Its access pattern is fairly unique — random writes and reads, and frequent overwrites, which means it could potentially wear out your typical prosumer SSD relatively quickly. But we’re not complete cheapskates, we’ll buy enterprise class SSDs, no?
So we go read some stuff and figure, hey, sounds like a great idea, right? Put TempDB on a local SSD, or better yet, a pair of SSDs in RAID-0 for pure performance (because this is a cluster, we’ve got HA already). We’ll reduce the load on the SAN I/O channels and make our overworked TempDB happier with lower latency and better throughput. Right?
In which we discover what could possibly go wrong.
Once the new drive(s) is(are) installed and “presented” to Windows (that’s my SysAdmin’s term), it’s fairly trivial to do the SQL configuration change — it does of course require a SQL service restart (or cluster failover). Code example, assuming your new drive is ‘T’:
use master; alter database tempdb modify file (name=tempdev, filename='T:\tempdb.mdf') alter database tempdb modify file (name=tempdb2, filename='T:\tempdb2.ndf') --etc...
You do of course have multiple TempDB data files, yes? Good.
Should we put
templog (TempDB’s transaction log) on the same drive as the TempDB data files, or put it on the same storage pool as the regular DBs’ TLogs? As usual, “it depends” — ask your favorite SQL gurus and do some testing.
Back on topic
We’ve made the change, we’ve done the cluster failover. TempDB is now running on our spankin’ new SSD. So we start monitoring performance metrics. Things like file I/O stats (from SQL DMV
sys.dm_io_virtual_file_stats), latency and waits (from our monitoring tools), and good ol’
But wait, what’s this? I/O stalls are higher? Write latency is higher?!? Perfmon agrees?
Write latency on the TempDB files was over 10x higher than it was when they were on the SAN (the performance tier, to be clear). The
file_stats DMV showed large increases in I/O stalls. Sad-trombone.
In which we have several theories
Then ensued various conversations and brainstorms among my colleagues.
Someone check the firmware/drivers!
Maybe it’s got the wrong block-size.
Well, it’s only 6Gbps SAS… maybe we should’ve sprung for the 12Gbps.
The write latencies went up by a factor of 10. I don’t think an improvement by a factor of 2 is going to win you any trophies.
Why didn’t we get an NVMe or M.2 one?
Because the damn blades don’t have those slots, goober.
Another interesting observation, and potentially the silver lining. Overall instance waits (wait stats), according to our monitoring tool, went down. That’s good news, right? Maybe. Does application performance & user experience corroborate it? Possibly! We’ll be observing the patient for another week or so.
Let’s turn to the community again to see what others have experience.
- Oh wait, it might not be such a fantastic idea after all.
- Yeah, Paul agrees. Twice.
- Crap, we should’ve done more research.
In which we eat some crow
And by “we” I mean “me”. Being the DBA and the primary proponent of the SSD addition, because I knew our workloads were very TempDB-heavy, I had to hang-tail and admit that the SAN gods won this round.
But wait, what about the fact that our wait stats are down? What about app/user experience? Valid arguments, I agree. That’s why we’re still observing. But I’m not optimistic, given the follow-up links above. We may utilize local SSDs for something else (index filegroups?) — but if those write latencies don’t improve, I’m concerned that it won’t help anybody.
In which I ask for your help
Yes, you! If you have ideas on what we did wrong, what we’re missing, or any other advice about getting the most “bang for the buck” out of a direct attached SSD on a converged-infrastructure Cisco UCS blade server platform with a VNX SAN, by all means, drop me a line. I’m all ears.