ST39103LC disk speed

[Jim Smith]

Hi, weve got an L class with 2 of these Seagate ST39103LC disks on it. Im trying to find out how fast these disks are. Seagates web page says the disk will do up to 40.5 Mbytes/s which is astonishingly fast. How can I check on my L class how fast these disks can perform under LVM ?

[Stefan Farrelly · Accepted solution]

To test these disks on HP-UX firstly use a raw dd to test the drives maximum speed;

time dd if=/dev/rdsk/cxxxxx of=/dev/null bs=1024k count=50
Take the results in seconds (eg. real 1.9) and divide it into 50 (50/1.9=26.3 Mbytes/s)
You should average a few runs to get a more reliable result.

To test it in LVM put the disk into a volume group and then test it using the same command but change if= to /dev/vgXX/rlvolXX and then again using lvolXX. You will see that using the rdsk or rlvol is much faster than using lvol as this bypasses the HP-UX cache, but this should give you a pretty good idea as to the disks performance under LVM (using a filesystem - non raw - I make it around 13 Mbytes/s) - im afraid nowhere near the 40.5 Mb/s quoted by Seagate!

[Andy Monks]

you could try just doing a dd.

something like :-

time dd if=/dev/rdsk/ of=/dev/null bs=1024k count=X

Where X is the number of MB you want to read.

That will do the raw disk speed. Then just use a lvol name instead of the raw device to see what it is under lvm.

Btw, you'll never acheive the maximum.

[Jim Smith]

Thanks for the replies.

Im not impressed with only 13 Mbyes/s when Seagate say up to 40! talk about misleading. Even raw only gives 26 Mbytes/s.

Is there anyway I can speed it up under LVM ?

[John Palmer]

No, but you could get some faster disks. Model ST39204LC's indicate a raw disk speed of about 34Mb/sec and LVM speed of 32.5Mb/sec.

This is on a new L2000/2.

Regards,

John

[Jim Smith]

John - these faster disks, shouldnt the raw speed be twice as fast as the lvol speed ? - as per the examples above (ie. dd...rdsk... and rlvol should be about twice as fast as lvol ?)

[Stefan Farrelly]

Theres not really anyay to make the disk perform faster under LVM. The best thing to do is any lvol on it stripe across multiple disks. ie. if you have 2 of these disks create your lvol using lvcreate -i -I across both of them and now when you do a;
time dd if=/dev/vgxx/lvolxx ... the result should be twice as fast as with 1 disk. The more disks and the more SCSI controllers you can stripe over the better!

There are some options under LVM which can make the lvol perform as if it was being accessed like a raw lvol (mount options - delaylog, mincache=direct, convosync=direct, nodatainlog) if youre using JFS/Online JFS. However, the rules under which LVM will access it raw are quite strict and only certain applications which are written to utilise this can do so (eg. Oracle can be). I havent yet seen a way to test this using simple commands.

[Jim Smith]

John,

another question. According to Seagate these disks use an Ultra160 controller, not UltraSCSI2. Does your L class have faster controllers ?

[John Palmer]

Jim,

Out of interest, I've tried a few combinations:-

Raw disk Raw LVM
ST39204LC > 34 33
ST39102LC 18 14.5
ST336704FC > 35 > 35
ST136403FC 27 27

It's interesting to note that Fibre Channel disks don't seem to exhibit any LVM overhead.

It's also interesting that the ST136403FC is markedly slower than the ST336704FC and as it was supplied as a replacement by HP, I'll be having a word with them about it!

Regards,

John

[John Palmer]

Jim,

One ST39204LC is on a 'SCSI C896 Ultra Wide Single Ended' card, the other a 'SCSI C875...'. Both are the same speed.

The ST39102LC's are on an older L server with the same cards.

Regards,

John