SANs are different Part 2

Well, we bought one. The EqualLogic PS6000XV, the 450GB flavour.

We're on our demo unit (a 5000-series) until the new one arrives, at which point we'll join the new one to our group and 'evacuate' the data over. Pretty nifty. That feature alone saved EQL the sale over the EMC AX-4 a company was trying to get us into. Sadly, they could not get us a demo unit in time, so we could not test in time. Ah well. We all really like the EQL solution, and features. Since we're close to Dell we got a pretty nice price as well.

Amusingly enough, the only complicated part of the SAN is the rack mounts - they are terrible! Or rather, the diagrams are...photocopied CAD diagrams at that.

We're using these specs to judge because the critical loads are SQL I/O, and the average request is around 50kb.

Our performance numbers looked something like this, keeping in mind this is the 10k PS5000. All are random reads, 64 threads to a depth of 128. Shown is the block size and max IOPS.

MD1000
- 64kb - 1456IOPS
- 8kb - 3058IOPS

PS5000E
- 64kb - 2640IOPS
- 8kb - 3393IOPS

PS5000X
- 64kb - 3601IOPS
- 8kb - 7389IOPS

According to what data we can find, the PS600XV should be at least 25% faster than the PS5000X due to 15k disks versus 10k, double the cache (2GB to 4GB), and an extra 1Gb port (for a total of 4Gb when using MPIO). I would guess the newer architecture would help as well, but I have no real basis for that.

I should stress that these numbers are positioned to make the MD1000 look like a dog compared to the 5000X, because in these instances it is. However, in the sequential department, the MD1000 was faster than the 5000X by a surprising amount. We discounted that because all our access would be VM, SQL, and Exchange - mainly random reads.

Comments

Popular posts from this blog

DFSR - eventid 4312 - replication just won't work

Fixing duplicate SPNs (service principal name)

Logstash to Nagios - alerting based on Windows Event ID