ProLiant Servers (ML,DL,SL)
Showing results for 
Search instead for 
Did you mean: 

Call for Smart Array benchmark data


Call for Smart Array benchmark data

I've been browsing the general net as well as the HP/Compaq sites for weeks now, trying to obtain some smart array performance data. Nothing there!

I can find all kinds of benchmark figures for ICP Vortex, Adaptec, Mylex, AMI etc. controllers, also lots of figures for IDE or SATA RAIDs, but for the Smart Array or the NetRaid controllers: nothing!

We have basically been a Compaq only shop over the last ten years with something like 20 servers and hundreds of desktops of all kinds running. In terms of RAID controllers there is a few 2SLs left over, lots of 3200 and some 5300. Still I've only very recently come across the fact, that all these wonderful "high performance" RAID controllers are real dogs. Even though we are running quite a few Windows releases (NT 4, NT4TSE and Win2k) and all kinds of drives with these systems, write performance on RAID 5 systems with 3 to 10 disks is usually in the order of 10% of the practical bus limit.
2.5MB/sec on EISA controllers
10MB/sec on PCI controllers
20MB/sec on 64bit-PCI-X controllers.

I've run various benchmarks, but the simplest ones have been NBENCH.EXE, which includes a disk benchmark and Intel's IOmeter, which is kind of hard to find these days (there is a version in SourceForge).

So if you have any kind a Smart Array controller and a minute of your time, would you please run any of these benchmarks and post the results (for a 999MB file)?
Adrian Ogden
Frequent Advisor

Re: Call for Smart Array benchmark data

You need to specify whether WBC is available or not as this makes a huge difference in performance. RAID 5 has never been very good without it.
Adrian Ogden
Frequent Advisor

Re: Call for Smart Array benchmark data

On a DL380R01, dual 1GHz, 2GB RAM, 3*34GB 10,000 RPM disks in RAID 5, built in Smart Array controller, no WBC, NT4.
Nbench 999MB file write: 2.78, read 23.66

Re: Call for Smart Array benchmark data

Yes write back cache is available, except perhaps on the 2SL.
Alle 3200 have the full 64/56 MB,
some 5300 128MB others the default 64/56MB.

I've tried all cache ratio settings, but usually leave things at 50%/50%. When I do the NBENCH run with 999MB I can see the cache taking the first MBs in one gulp, then it slows down to whatever the controller is able to sustain (10% of what I'd expect).

BTW RAID-0 values are almost as bad.
John Marinov
Frequent Advisor

Re: Call for Smart Array benchmark data

Here's the trap. WBC is not available without battery backup. The DL380G1 can't get it. Neither can the 8500. The later DL380s have the 5i battery option.

Same problem for all the "integrated" smartarray controllers.

Makes sense when you think about it. The only way to guarantee the write will succeed is to make sure the cache is battery backed (is this why the drive cache is turned off?).

2SL is not battery backed. 2DH is.

If you tell us what sort of equipment you're baselining, and where did you get your benchmarks and software, I'll organise some testing here and post the results.

Re: Call for Smart Array benchmark data

True a dreadfull performance, we own some ML370 with 3200 controllers with exact same problem !

Re: Call for Smart Array benchmark data

Just to clarify...
For the purpose of this call for benchmarks our baseline is this:

a) SmartArray 3200 (32-bit PCI) with 64MB battery-backed write back cache (configured at 50%/50% read/write ratio) in dual processor ML370 G1 at 933 or 1000 MHz

b) SmartArray 5300 (64-bit PCI) with 64 or 128 MB battery-backed write back cache (also configured at 50%/50% read/write ratio) in dual processor ML530 with 2.4GHz.

All Disks are 36GB10K, 36GB15K or 72GB10k and are configured as RAID-5 with six disks per server.
Chad Boulanger
Occasional Visitor

Re: Call for Smart Array benchmark data

This thread is 6 months old but I would be interested in seeing the data asked for. Has anyone run the benchmarks? Any thoughts on how the Smart Array cards do when compared with those from LSI, Adaptec, etc?
Those who give up liberty for the sake of security deserve neither liberty nor security. Ben Franklin.

Re: Call for Smart Array benchmark data

I have a write performance problem with a Proliant DL 380. We have write intensive real time software and we need an excellent I/O performance.

We have a Proliant DL 380 and we got a max of 83 records/sec while in a COMPACT Evo with a 40 GB IDE HDD we got more than 500 records/sec.

Both systems have Windows2000 SP4 and SQLServer 2000 SP3. The Proliant Server has a Smart Array 5i RAID controller.

I've run the NBENCH.EXE disk benchmark in the server and my desktop COMPAQ Evo PC. We got the following results:

Compaq Evo:
file size: 999.0 MB
write: 40.24
read: 41.84

PROLIANT DL 380 Server:
file size: 999.0 MB
write: 12.61
read: 66.39

RAID 0 36GB single disk:
file size: 999.0 MB
write: 11.73
read: 41.98

Mogens Kjaer
Frequent Advisor

Re: Call for Smart Array benchmark data

I'm trying to set up a ML370 with a Smart Array 5312, with 8x146G disks i RAID 5.

I was very much amazed to learn that software
RAID 5 in Linux was more than twice as fast for
reading and writing than the SA 5312.

Writing a 10G file, and reading it afterwards,
gave me the following numbers:

SA 5312: Write 31Mbytes/sec, read 81 Mbytes/sec
Software RAID5: Write 84 Mbytes/sec, read 186 Mbytes/sec

The SA has 128M cache (battery backup)
split 50/50, stripe size 16K

The software RAID was set up with 7 disks only,
as Linux can't boot from a software RAID 5. The
discs were attached to the on-board adaptec controller for the software raid.
Occasional Visitor

Re: Call for Smart Array benchmark data

I am experiencing terribly slow performance on a DL380 with a 5i card (no battery)... 9MBytes/Second write and 82Mbytes/sec read of 999Mbyte file using NBENCH (single thread) results for a 3 disk RAID5 setup.

As you can see, terrible. What is going on?

Adam Thomas_2
Occasional Visitor

Re: Call for Smart Array benchmark data

Better late than never.... Hopefully attached to this posting is a short spreadhsheet with some stats I have collected. I actually collected them before reading this posting so the file size is only 50MB. I think the stats are probably still OK given the performance of my DL380 with RAID5 is near identical to a posting on here already.

Some of the servers I tested are running in Production so hammering them with a large file isn't an option, and of course some of the stats will be affected down somewhat due to contention with the other activity. I did pick quiet times to run them through.