EVA3000 and MA8000 IOPS

[陳文賢]

Sir :

I wnat to find the differenence of performance between EVA3000 and MA8000. Do anybody know the IOPS about EVA3000 and MA8000 ?

[Uwe Zessin]

That depends on the number of disks and their configuration. Or would you like to know what the controllers theoretically can do?

[陳文賢]

sir :

Yes,I want to know IOPS per controller in MA8000 and EVA3000 .

[Uwe Zessin]

The specifications claim 'over 24kIOPS' per HSG80 controller pair, but that's not realistic.

An MA8000 can have up to 42 disks - that would mean about 570 I/Os per disk.

For the EVA-5000, the claim is 'Up to 141K IOPS and up to 700MB/s throughput per Controller Pair'.

With 240 disk drives this would be 587 I/Os per disk.

--

One might be able to get these numbers with special benchmarks, tweaking some controller parameters and tricky use of the controller cache, but those numbers just show how powerfull the controller hardware and software is. In real life you will _never_ see these numbers.

[David Ell]

That is impossible.. As a rule of thumb you usually use anywhere from 110 - 150 IOs per spindle (disk)

[Uwe Zessin]

Carefully re-read my statement and you should understand that I wasn't suggesting a single disk drive could do 500+ I/Os in a typical real life situation.

[Mahesh Kumar Malik]

Hi Woody

1. The MA8000 has two fibre channel ports per storage controller or redundant controller pair. Each controller-to-host interface provides up to 100MB/sec of data throughput for a maximum of 200 Mb/sec available capacity in each controller

2.EVA3000 Fibre Channel host connections provide up to 200 MB bandwidth for each path. Each controller has two Fibre Channel host ports (four ports in a redundant pair of controllers) assuring the availability of bandwidth for the most demanding applications. In addition, up to 2 Gb of cache per controller pair ensures high performance. Mirrored write caching capability maintains optimal availability while assuring data integrity in the event of a failure.

Regards
Mahesh

[generic_1]

Id pick the 3000, unless you feels you have it grossly overloaded. Its easier to take care of and manage, and faster. I also think its more reliable, but its important to have monitoring on that puppy. Critically failed disks bring EVA's to their knees at times. Its important to not dilly dally with bad drives on these :).

[Uwe Zessin]

Mahesh:

1. The _bandwith_ on the MA8000 controller ports is 100 MegaBytes / sec., but it is highly unlikely you will get 100MB/s out of a single controller even when using 2 ports.

2. So far EVA has _always_ been sold with two controllers. Only the EVA8000 can be equipped with up to 2 GigaByte data cache memory (not counting the so-called 'policy memory') per controller.

[Bostjan Kosi]

Gents, MA8000 fully loaded can give a max of 21,000 IO/sec (.5k) and a EVA 5000 (168 disks - not fully loaded) gives a max of 162,000 IO/sec (0.5k). These numbers ar not relaistic, since the write blocks are 0.5K in size and also the cache gives a lot to the performace of EVA. Realisticly ( without cache at all) MA8000 can give 10000 iops and EVA can produce 36000 IOPS. The top performance I have measured at a customer site on EVA5000 with 140 HDD (36, 72 and 146GB / 10 and 15K) was 25000 IOPS (peak). The customer is running large ORACLE and some MS SQL servers, so this number is realistical (real trafic not benchmarking). In your case of course you have EVA 3000 limited with the amount of disks, so munbers like these are not achievable even in teory... EVA works very well with the CACHE... You can count of max 110 to 150 iops per HDD..so just multiply...plus add some good cache performace and there you go...

BR
B.

[Cass Witkowski]

We did some testing with HSG80s and EVA3000.

The HSG80s had 32 9 GB 15K rpm disks each one was a set up as a JBOD with no cache enabled. (no read, no writeback and no read-ahead). We used TESTDEV on OpenVMS system. We did random 1KB reads on the disks. The maxium we could get through the HSG80 pair was about 4,700s I/O per second. Since this was with cache turned off this was I/Os from the disks.

On the EVA we had 48 36 GB disks which we created 32 vraid 0 disk of 9 GB in size. We ran the test and got over 30,000 I/Os per second. Again cache was disabled.

On a MSA1000 we did a similar test and saw about 10,000 I/Os per second

YMMV

Regards
Cass