DS 1511+ with Vertex 4 SSD- IOPS- My Results!

[jawuk]

Hi all

Sometime ago, i posted a message asking whether i could expect to see a good results for IOPS and latency performance using SSD drives in the Synology 1511+. As we all know, SSD is very well known for superb IOPS performance over mechanical drives, so for applications which demand high IOPS, that usually would require an elaborate SAS SCSI SAN can be achieved using lower end hardware, and a simple high reliability SSD (or two).

I have recently installed an OCZ Vertex 4 256mb ssd in my Synology 1511+ NAS, to use as my block level iscsi data store for my VM Ware Cluster.

Anyway, this drive is capable, connected natively to a system with SATAIII, at achieving 445mb read and 457mb sequential write, and iops with 4k64thread test on AS SSD 86,000 read iops and 76,000 write (http://www.guru3d.com/article/ocz-verte ... -review/14) .

Clearly, installing this in a NAS with SATAII is ofcourse going to impact these metrics, as is, connecting over a single 1GBps connection, but still, i should still see a vast improvement in IOPS, and max out my 1GBps connection.

Well, certainly, the results are interesting

Over a single 1GBps connection, I am getting a more modest 102mb/s read and 110 mb/s write, and 9269 read iops and 8842 write with the 4k64thread test on AS SSD and 0.6ms read and 0.5ms write latency



IOPS results are very impressive over a mechanical drive. It would take ALOT of VMS, or exchange servers to come anywhere near that demand. To put that in persepctive a single SATA 7200rpm HD will achieve 75 IOPS, and a 10,000rpm SAS drive 175 IOPS.

To put that in perspective: -

An IBM DS5300 SAN with 16x 15k 3.5" drives will achieve 3500IOPS (100% random 4k) and have a latency of 4.2ms. That SAN would be well over £6,000 including disks.


While i am pleased, i am keen to try and improve my IOPS performance even more, and also, understand why my IOPS performance should not be higher still.

Where would my bottlenecks be do you think? While i know my network interface could affect throughput (i am getting around 118mb/s read/write), i did not think that it would necessarily be limiting IOPS.

Any ideas?

I have been using AS SSD tool for benchmarking.

Regards

James

[badedyr]

Hi James,

Can you detail your setup? Im currently having big issues with getting anymore than 800-1000 IOPS read or write 4kb using any kind of benchmarking (AS SSD, IOmeter etc.) from NFS or iSCSI on a DS3611 box.
Running ESXi 5.0 U1 and seeing 200+ MS latency average on both read and write when running approx 1000iops 4kb random against Intel SSD disks.

[Eideen]

To get more then 110MB/s you need a switch that support "IEEE 802.3ad" and your pc need to have 2 network cards that need to support it to. but it will still cap at ~220MB/s

To get anything more you need a NAS/switch/PC that support fiber optic.

Eideen

[gabi_cavaller]

That seems like a ridiculous amount of IOPS, are you sure that is the case?

That is a very large amount.

[jameslol]

That is super huge - but not unbelievable.. Maybe one of the popular iometer icfs could be a good second test. Benchmarkreviews have the one they use in all their tests here: http://benchmarkreviews.com/images/arti ... W_30QD.icf

Eideen: Probably wouldn't use 802.3ad - It will still not exceed 1gbit per single connection (fine for distributed connections from multiple clients). The common method is MPIO iSCSI.

I've got a ds412 for testing with MPIO iSCSI with Vsphere right now. Will be getting a 2200 or 3400 device also, if all goes well. I would be interested to see if your IOPS reading doubles with MPIO.

If you do give MPIO iSCSI a shot, you might want to check out my thread here to see if this also effects you viewtopic.php?f=145&t=54024

Cheers
James

[jawuk]

Here you go guys! Here are some screengrabs





J