EVA; i/o versus MB/sec

[Wesley Staal]

We have an EVA5000 with a FC scsi diskgroup which contains 40 x 146GB 10k disks. According to our storage provider the system has a maximum throughput of 150MB/sec write and a 525Mb/sec read performance. Adding more disks will raise the number of write i/o per sec, but it will not raise the numberof write Mb/sec the diskgroup can process. Can anyone explain the relation between the io/sec and the Mb/sec in an EVA system.

[Uwe Zessin · Accepted solution]

Those 'great looking' numbers are created by special test programs which have nothing to do with reality.

High I/Os are created by doing small I/Os to the same disk location so that the system just operates on the cache. High MB/s numbers are created using large sequential I/Os. Both are great to test the controller's performance (CPU, bus bandwidth), but not apply to reality.

Many installations I am aware of use a single path for data traffic and the bandwith of that one is sufficient.

[Hein van den Heuvel]

What Uwe said!

But let's add this:

10K RPM = 166 Revolutions per second.
So that is pretty much the maximum IO rate per second... per disk.
With 40 disks that gives 6000+ IO/sec
The EVA5000 has the compute/control power to drive say 20,000 IO/sec (dunno, must be at least that), so there is still room to spare. And adding disks may help the final performance.

For the bandwidth the above system wide number sound reasonable. On the writes the data has to go to cache and to two drives (redundancy) so you'll run into controller backplace limitations. The reads are often (fibre) channel restrictions. Well each disk can readilly deliver more then 20MB/sec. So just 25 disks will readily reach the max read throughput, you have 40. Those can deliver 2x or more MB/sec than the controller/Fibres can swallow. So more disks will not help bandwidth alone,
Now if the application does medium sized IO (8kb - 32kb) and has some randomness then you may well need more the 40 drives to get to see the MB/sec desired.

Makes sense?
Hth,
Hein van den Heuvel
HvdH Performance Consulting

[Wesley Staal]

Guys,

Thanks for the reply. The 150Mb/sec is actualy reached while copying .jpeg files from about 2MB each using several fileservers. One fileserver can write jpeg data at about 40Mb/sec. Does this mean that while using more than 4 servers at the same time the EVA will be a bottleneck?