speed write to disk

[Berend Tooms]

Hello

I use a ML350 (Novell 6,5) with a SA641 scsi raidcontroller with 4 disks 36 Mb. When I write to disk the speed is max. 1200 kb/sec by small blocks (512) this speed is getting higher with greater blocks up to 8.000 kb/sec using blocks from 10.000 kb. We are using a application thats writes small blockes and this 1200 is nog fast enough. How can I make this go faster, if possible?

Thanks in advance

Berend

P.S. This thread has been moved from ProLiant Storage Systems to ProLiant Servers (ML,DL,SL). - Hp Forum Moderator

[Peter Godron]

Berend,
can't you buffer the output to memory until you have filled larger blocks?

[Berend Tooms]

Peter,

This is a database application we are using. Writing and reading is real time by several users. So saving for bigger blocks is not an option. It is not a problem in my network. When I try a crosscable on the server, it is still the same results.
I think my question is: is this 1200 kb/sec a normal result? Or?

[Peter Godron]

Berend,
according to the tech guide (attached) the bandwidth is there.
What sort of RAID setup have you got?

[Vincent Fleming · Accepted solution]

1200 KB/s at 512-byte blocks is darn good for 4 drives! That's 600 IOPS per drive! (They're rated at 120-160 IOPS per drive)

I'm surprised that you are getting that much...

There are 3 things you can do to increase your performance:

1. Add the 128MB battery-backed-cache option (if you haven't already)
2. Add more disks (I think you're limited to 6, but you should be able to add 2)
3. Add an external disk array. This could be another SmartArray with an MSA30 shelf, an MSA1000 or MSA1500, or an EVA4000. This option allows you to use many more disks.

Good luck,

Vince

[John Kufrovich]

Berend
For DB applications you shouldn't concentrate on the MB/s but instead look at IOPS.

[Berend Tooms]

Hello,

These are things that I can use in my contact with my softwarehouse! This is all comming out by tests that has been done by a softwarehouse for using new software.
I'm using a raid 5 with 1 spare, by the way.
I thought that the problem could be the NIC (10/100/1000,server -auto- vs 100,computer or switch) or memory (1,5 Gb), but that seems not to be the problem. When I have listened well, there is no problem. By the way, the performance of this server is perfect, so....... why change a winning team. I think not, untill........

Thanks all. I'll let you know when something changes.

[Leif Halvarsson_2]

Hi,
You can increase write performance if you change the Raid level fom Raid5 to Raid 10 as Raid10 has less write overhead then Raid5 ( Raid5=4 IO, Raid10=2 IO ).

[Simon Setina]

Hi,

for increasing IO, you should think about Solid State Disks.

They are fast. With SSD you can achive more than 70.000+ IOPS with no problem.

Simon

[Vincent Fleming]

Simon - he has an ML350.

It's not practical to put a $50,000 SSD on a $5,000 server...

[Simon Setina]

Vincent,

Yes SSD cost a lot (from 30,000 to sky is the limit $)if using external FC box. But PCI/PCI-X SSD are much much cheaper.

You also propose Berend an EVA box.
"3. Add an external disk array. This could be another SmartArray with an MSA30 shelf, an MSA1000 or MSA1500, or an EVA4000. This option allows you to use many more disks"

At the end. With smart optimization of database, using a lot of small 15K rpm disk in RAID1 not 5 a significant improving can be expected.


Simon

[Berend Tooms]

Hello everyone,

I have done everything that is reasonable. A BBWC 128Mb, 2 Gb extra memory (now 3,5 Gb), but nothing has changed. Has somebody any idea to tweak this machine? A colleague has mroe or less the same configuration, but uses W2003 and writes almost 5000 and reads 5 times more. What are we doing wrong???

Berend

[david murray]

How do you measure the IOPS or otherwise benchmark the drive?