VA7100 -> EVA4000

[Kurz]

Hi,

we are planning to replace our VA7100. It seems to me so far that the EVA4000 would be most suitable.

If you assume that both arrays have the same disk speeds and sizes, same raid level, same host, what is the performance gain to expect?
In terms of MB/sec read and write. SAN connection speed will probably not be the limiting factor?

I can't find any information on this in the Quickspecs or data sheet.

Any information or test results would be greatly appreciated.

regards,

M. Kurz

[Luk Vandenbussche]

I replace a VA7100 by an EVA3000 yet.
Works fine.

I think you find an answer for your questions on

http://h18006.www1.hp.com/products/storageworks/eva4000/index.html

[Vincent Fleming]

The highly technical answer to your question is: It Depends.

The EVA has more powerful CPUs, and doesn't have AutoRAID, so there's a little less for the controllers to do.

This translates into higher throughput and lower latency in general.

How much of a difference you will see will vary greatly depending on your configuration and access patterns. Your Hosts' performance is also a factor - is it powerful enough to drive the EVA harder than the VA?

For example, if you have 10 or fewer drives, you probably won't see a very remarkable difference in performance between the VA and EVA because both would be drive-constrained.

As you add more drives, however, you will see an increasing performance difference, particularly with RAID5 in the EVA vs RAID5-DP (aka RAID6) in the VA.

I hope this helps,

Vince

[Ted Buis]

The EVA4000 is a faster engine than the VA7100, however, if you are really hitting the array hard, you can be limited by the "back end" performance of the number of spindles that you have in the unit. I would focus on increasing the number of spindles in your EVA4000 to make sure you are not back end limited. If you are doing many small random I/Os, then each physical disk can only do 100-200 I/Os per second. If you are hitting it hard with writes, the array's cache fills and it must wait until data is written to disk, thus limiting performance to the number of physical disks regardless of how fast the controllers are.

[AwadheshPandey]

EVA is much better than VA,
pls check this thread to know diff abt VA and EVA,
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=972020
u can visit hp site to know more abt EVA4000

Awadhesh

[Kurz]

Thanks for your answers. I admit I still have no idea what performance gain in % to expect.

Aren't there any test results available?

[Vincent Fleming]

Kurz,

That's the point - you can't expect any particular % of performance difference.

If you can give us a configuration for the VA and the EVA, we could make an educated guess, but that's about all it would be - a guess.

Regards,

Vince

[Joshua Scott · Accepted solution]

A quick comparison:

VA:
512MB of cache
10K or 15K 100MB/sec Drives
100MB/Sec total bandwidth
Maximum of 15 Drives

EVA:
3072MB of cache
15K 400MB/sec Drives
800MB/Sec total Bandwidth
Maximum of 56 Drives.

The EVA will give a performance boost of 30 - 200% depending on what your usage is.

Josh

[Ted Buis]

I'm not sure where Joshua gets his numbers. None of these disks can sustain the MB/sec rates he quotes through seeks and track switches. My guess is that he is looking at the 1Gb/sec FC on the VA versus 2Gb/sec FC on the EVA, but both FC drives are dual ported, so maximum bus bandwidth would be roughly 200 and 400MB/sec with a single enclosure. Since the VA had 15 disks per enclosure and the EVA has 14 disks per enclosure, then if you compared 15 drives on each you have two disk enclosures for the EVA it would quadruple the bandwidth to the backend. The EVA with two enclosures will be faster for large sequential transfers, but for typical OLTP applications I would focus on increasing the number of physical spindles and take 28 10K drives over 14 15K drives. Naturally, 28 15K drives would be faster, but also more expensive.

[Ted Buis]

To follow up on what Vincent was saying, if you had 8 drives of 10K of the same capacity in both the VA7100 and EVA4000, I would expect any heavy mixed read/write random I/O (OLTP) application to barely above 0% improvement, because the disks are the limiting factor, not the controllers, buses or cache. The I/O potential capability of these arrays increases as you add disk drives to the unit.