Bonnie Benchmark, Solaris and EVA

[Bob Pace]

Does anyone have any benchmark data using the "Bonnie Benchmark" running on Solaris against an EVA storage array.

I am trying to help one of my customers who is trying to tell him how "poor" his EVA is performing using this benchmark.

It is a very political issue because the consultant is pushing an EMC box as a solution and I am trying to come up with external data samples.

[Craig_83]

Hi Bob,

I haven't seen any results with the Bonnie Benchmark... however, depending on how the EVA was configures could sure impact the results.

Have you seen the SPC Benchmark results? The URL is http://www.storageperformance.org/results/#cat1

Good luck,
Craig

[Paul Henderson_2]

Bob,

I am running three Solaris V240 servers connecting to an EVA5000. We have a 200+ system linux cluster that is NFS mounting filesystems from these servers in a high performance technical computing environment, and doing so over regular 100MB networks. The cluster runs at 100% load 24x7 doing fusion research computations.

We migrated to the EVA and Solaris servers due to very poor performance from a RAID array controlled by Linux boxes. Since then we have experienced:

- zero NFS server timeouts (we used to have hundreds per cluster system)
- no outages
- 475% performance improvement on writes
- 400% performance improvement on reads

IOW, we are thrilled.

We looked at EMC, and it was a dog compared to EVA. The EVA is truly a gem in HP's portfolio!

[Bob Pace]

Thanks for your input

Do you have any performance numbers for a single unix server writing to a single LUN on the EVA. We are see numbers that seem very low..

Writes in the 35 to 45 MB/sec range.

This is to a VRAID5 lun in a 120+ disk disk group.

Big solaris box running securepath 3.0c and 2 HBA's

[Craig_83]

Bob,

What makes up the fabric and what type of Solaris server are you using?

In a Sun performance class, they recommended increasing the following parameters in /etc/system when you have high performance stoarge:

set ufs:ufs_HW=16777216
set ufs:ufs_LW=8388608

ufs_HW is the number of bytes outstanding on a single file barrier value. If this value is exceeded, the write will be deferred by putting the thread to sleep.

ufs_LW is the low water mark where threads that have been waiting will awaken and issue their write request.

Our Sunfire 4800's and 480's routinely see 70-80MB/s for writes. This is on a EVA-5000 with two disk groups of 84 72GB 10K RPM drives.

What VCS are you running?

Craig

[Bob Pace]

It's a pretty heavily zoned fabric with Brocade switches.

Mix of Sun and Windows boxes. Windows boxes writing in the 50 to 60 MB/sec range.

The server is a Sun 6800 with something like 60GB of memory and 16 CPU's, 2 HBA's

I believe it's the latest version of VCS because they are on a pretty high level HP maint. plan

[Craig_83]

Bob,

Another thing to consider is the first write effect. The EVA keeps track of which virtual disk blocks have been written. This usually is not noticed during normal use...but will distort benchmark write results. To eliminate do the following on Solaris:
dd if=/dev/zero of=/dev/rdsk/cxtxdxsx bs=64k

Craig