HPE Storage Users Group

Good Morning!

I am trying to figure out a performance issue that we are seeing on our V800 with 1.2PB. We are 3.1.2 mu3. I have some RHEL 5 boxes that are having throughput issues. Running a dd command at 1M or greater, we are getting 450-500MB/s. When we drop to 512k or 256k we are getting about 40MB/s. This is not happening on Window Servers.

The server has 2x2TB luns presented and it exhibits on both. I also two other Linux server with the same results. This server does logging and takes in device logs and parses them out so the files range from 1k to 1M usually.

One question would be could partition alignment cause this much of a problem?

TIA,

JD

Be careful trying to do comparisons with dd as it's not a good benchmark tool for SAN storage like 3PAR, it's sequential and single threaded, so typically at smaller sizes your results will be poor. If you look at the array during such a run it will usually be spending most of it's time just waiting for dd to issue the next I/O.

I strongly suggest System Reporter as it will give you very detailed info on what the array thinks it is doing down to port, disk, volume, etc. You can see service times, throughput, etc.

I dont have too many options with RHEL unless someone can give me some other ideas on benchmarking tools to run from cli.

The problem seems not to be the SAN as the reports I am generating from System Reporter shows little IOPs or bandwidth.

I need to be able to generate some consistent I/O to prove to the customer that the SAN is not the issue. We use IOmeter on Windows but don't have the luxury on RHEL since we dont have gui installed. On windows just on a simple test at 256k 100%write/read and 8 IO requests we pushed close to 40k IOPs and 500MB/s plus.

The application developers continue to point fingers at the SAN.

UGH!

jd

I think Cleanur is right, those big arrays are designed for lots of parallel workloads. iometer is difficult as well, you really need to know how to set up the test, or you just measure the cache-performance or whatsoever.

I recently saw a slide/graph for a 7400. On the x-axis was the numer of simultaneous threads, the y-axis was IOPS I think. The performacne of the array was very poor at 256 or less, and started to be acceptably good at about 768+ simultaneous threads.

ioMeter is a good benchmarking option.

yeah use IOMeter on the Linux guests as slaves to the the GUI on a windows box. The IO is still generated on the Linux guests.