300GB Disks 4 x slower than 72GB disks

[Ian Dennison_1]

Help! I have a bunch of P.H.Bs walking around spouting how 72GB EVA disks are 4 x faster than 300GB EVA disks. Does anyone have a muppets guide to disk arrays? I have tried to explain seek times and block sizes and get blank looks all around.

Share and Enjoy! Ian

[Víctor Cespón · Accepted solution]

Humm, do you mean disks with the same RPM and the same number of disks? Then no, you should get similar speeds.

Another thing is if they are talking about 1 TB made with 72 GB disks versus 1 TB made with 300 GB disks. 4 times more disks give you 4 times more speed, obviously.

Check http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01671044

[Ian Dennison_1]

Thanks for those figures. If I understand correctly, both 300GB and 72GB 10K fibre disks get the same IO/sec for each Vraid type?

Is there any benefit in converting to RAID1 for a supposed performance increase? Also, would anyone know the max IO/sec for each EVA 5000 HSV110 Controller?

Share and Enjoy! Ian

[Torsten.]

Measure performance is always difficult because there are so much dependencies ...

The quickspecs only say "up to":

Sustained I/O and MB Throughput Up to 141K IOPS and up to 700MB/s throughput per Controller Pair

http://h18000.www1.hp.com/products/quickspecs/11006_div/11006_div.html


IMHO larger drives are usually a bit slower, but 4 times? How did you measure this?

[Ian Dennison_1]

The presumption I was given by my managers was based on disk size (4 x as big = 4 x as slow).

I imagine the block size is 4 x as large as well, which might work the EVA cache a little more but should provide roughly the same IOPS and data throughput.

Share and Enjoy! Ian

[Torsten.]

"4 x as big = 4 x as slow"

... now compare a 18GB and a 1TB drive ...


;-))

[Víctor Cespón]

Some answers to questions above:

When calculating I/Os per second the only things that matter is the type of drive and it's rpm. This is because the limit for random I/Os is the seek time of the drive heads (compare a mechanical disk to a SSD).

The sequential transfer rates increase with size since the amount of data that passes under the head in each revolution is bigger.

The limits of the EVA controllers are irrelevant, since you'll only reach that with a fully loaded EVA and in some non-real life cases. The limit are always the disks.

Yes, vdisks in RAID 1 are faster than in RAID 5, but take more space obviously.

[Uwe Zessin]

Disk drives are typically not accessed sequentially.
- sorry, could not resist.

The differences are in the number of platters, seek speed - in the old days some drives had multiple heads per surface to reduce seek distance, bit density and geometry. But if you get blank looks...


The sales man inside me would ignore all this technobabble and happily sell many EVAs with many small disk drives ;-)

[IBaltay]

Hi,
From a pure spindle speed side, the drives are almost identical. But the 300GB Disks contains 4 times more vdisks which creates an illusion that the 300GB disks are much slower

[Rob Leadbeater]

Hi,

Do you currently have both size disks in your EVA ? Are the disks in different groups, with different quantities of disks in each ?

The performance of a disk group is dependant on the number of disks within it, so if you have say a group with 32 x 72GB disks and a group with 8 x 300GB disks, you'd getter better performance from the former. Would it be 4x faster ? Doubtful...

Cheers,

Rob