iSCSI performance

Take a look at ReadyNAS Pro Series Performance numbers.

Dell performance numbers would be found elsewhere.

digitaldentistry wrote:
How is the performance of Pro2 in RAID 1 compared to say SATA drives on a Dell PowerEdge T310 running in RAID 1? I have tried doing some research online but can't find any performance testing results and data. This is in an environment where there are many small file reads and writes on a 200 to 300GB data set.

Well, I just tested a T310 with a Dell SAS 6/ir controller and two 500GB SAS Seagate Barracuda ES.2 drives (7200rpm) and it topped out at 120 MB/sec in linear transfer speed. That's pretty much the speed of the disks.

iSCSI performance on the Pro, well I don't have the Pro, but I have the Ultra 2. There is some serious overhead in the iSCSI department, I think, and it shows in the official numbers as well; a single drive which will give me 100MB/sec attached directly to SATA will deliver 60-70MB/sec in the Ultra 2 over iSCSI. Access time (seeks) is not affected at all. That's good enough for lots of things of course, but personally I don't think these devices are up there with a real server -- yet (they will be though, and they are better for a few specialized things).

That SAS 6/ir isn't very impressive either, a real RAID controller with a real array (meaning, more spindles in level 5/6 or even 10) will easily deliver several hundred MB/sec... but then the problem becomes more "what to do with all that data"? Which interface to fill it with?

I'd say a T310 in pretty much any configuration (even using Windows built-in software RAID) will absolutely smoke a Pro 2, but price, power requirements, cooling requirements, ability to easily replicate to another identical NAS (two units will still not even approach the price of a server) is in the NAS' favor.

Horses for courses and all that. 8)

Do note that the Ultra 2 has a slower CPU than the one in the Pro 2.

mdgm wrote:
Do note that the Ultra 2 has a slower CPU than the one in the Pro 2.

Certainly, but even the official performance numbers you link to shows a serious slowdown when using iSCSI. Also, in several of the benchmarks WRITE performance is higher than READ performance, which of course looks anomalous! So, something is wrong with the iSCSI thing (and the problem might even lie in the choice of IET, which I have come to understand is pretty much the worst choice in iSCSI but probably was chosen for a reason, kernel-wise or whatnot).

So far, I have investigated different initiators, but that had no effect at all, next step is probably changing from FILEIO to BLOCKIO on a target to see if that affects anything. I'll report back.

IET? We haven't used that in like a year now, unless you specifically did something to switch back to IET anything after 4.2.15 wouldn't use it by default.

Grievous wrote:
IET? We haven't used that in like a year now, unless you specifically did something to switch back to IET anything after 4.2.15 wouldn't use it by default.

Oh, it's LIO of course. Sorry. It sure looked like IET, the system even rewrites ietd.conf (always changing to fileio if I try to use blockio) automatically at start of the iSCSI daemon (from Frontview). That explains some things. :lol:

However, top doesn't show the CPU to be maxed out when hammering the target. Are there any hints for iSCSI performance tuning?