Disk Performance Load Tool, does such a beast exist?

[Tim D Fulford]

Hi

I'm looking for a disk performance tool, well loading tool. That is one that writes both sequentially & random IO's. I'll probably monitor the output with glance. I do not want to just use dd as it is just sequential.

I'm think I saw a reply saying there was one, which I duly noted, then forgot! I think Clay or JRF gave the reply, but I can't find it! (they have given ALOT of replies)

Here's hoping

Tim

[Stefan Farrelly · Accepted solution]

You want a tool like bonnie or iozone. Youll have to do a search for them though.

Heres a good site which mentions them;

http://www.nswc.navy.mil/cosip/may98/cots0598-1.shtml

[Santosh Nair_1]

Here's another link with another batch of disk benchmark utilities:

http://www.acnc.com/benchmarks.html

check out Bonnie++ and IOzone II.

-Santosh

[Tim D Fulford]

Many thanks for the above. Are there any HP specific tools?

I noticed that bonnie++ mentioned that it was usefulf for 32bit OSes. We currently want to look at stressing an fc60 with an N class server, which is 64bit? Or, is this irrelevant as the code willl be compiled on a 64bit compiler?

Has anyone done anything similar? If so I would like to hear how you did it.

Cheers

Tim

[Stefan Farrelly]

You can only use the standard HP tools, glance, PerfView, iostat, sar etc. PerfView has some excellent low level details on disk performance.

As for your FC60, theyre pretty standard, HP even have a document on the fastest way to configure them. We tested it here, worked a treat, got a massive 160+ MB/s over both fiber channels using RAID0+1, 4 disks per LUN, default stripe size on the SC10's, lvmstriping at 64k also over both channels. Cant get any faster!

[Tim D Fulford]

Stefan,

Surely not 160MS/s! That means your getting 40MS/s/disk! V fast disks! I also assume it was sequential part of the write.

Tim

[Stefan Farrelly]

Why not 160 MB/s ? Each fibre channel can do a max or around 85 so if you stripe an lvol over both channels thats a theoretical max of 170.

After we striped we tested using raw dd on the rvol and rdsk, same speed. Didnt test writes, but no reason why you cant. Create the lvol and use prealloc to create a huge file then copy it to the lvol, and try dd'ing it to the raw lvol (will destroy the lvol, newfs it again after).

[Tim D Fulford]

Stefan,

I'm not saying it is impossible, it's just my disks are 18GB 10krpm & do 15MB/s. If there is another flavour of disk that does 40MB/s great, I'm genuinely impressed.

Cheers

Tim

[Stefan Farrelly]

Hi Tim,

your 18GB 10krpm drives only do 15 MB/s ? thats pretty poor. Weve got internal 18GB Seagate ST318404LC drives (on N/L's with Ultra SCSI) which I can get up to 35 MB/s by doing;

time dd if=/dev/rdsk/c2t6d0 of=/dev/null bs=1024k count=50
50+0 records in
50+0 records out
real 1.4
user 0.0
sys 0.0

Which is 50 (MB) /1.4 secs = 35.7 MB/s.

Do you know what model your drives are ? and are you testing using the raw device ? (always much faster - for theoretical testing though).

[Tim D Fulford]

Stefan

Hmmm... I'm using the same spec of disk in vg00 & on fc60. I'm confused, my H/W experts say that my disk can do 15MB/s (advertised spec), they say that 18GB disks that can do up to 45MB/s individually (I've just been told). When I look at HP storage they advertise all of the disks in sc10/fc10/hvd10 as having 160MB/s, which must be wrong (in an fc60 I could see a LUN getting 160MB/s but a single disk?).

Please feel free to fill in the gaps

Tim