expected iops from ts879u-rp

[ncvbit]

With raid 5+hs (8 Hitachi 2tb ultrastars) and intel x540-t2 10Gbe?

Also, if I have 2 qnaps can I link them together for fail over or load balancing to increase iops?

Thanks,

Brandon

[sl1000]

Every sata drive does at best 70 iops. But from the 8 drives you loose one for hotspare, and the performance of one because of the raid-5 parity. So don't expect much more then about 6 x 70 = 420 iops sustained. Caching can help to increase this value if it is the right io, but i wouldn't count on it to much.

[sl1000]

Linking is not possible. Daisy chaining with iscsi might be, but i would prefer to connect multiple qnaps directly to the host.

[schumaku]

From TS-879U-RP Features - Highlights:

10 GbE ready reaching over 1,000 MB/s and 100,000 IOPS

[sl1000]

100.000 iops... yeah sure... with 8 SSD's in raid-0, 100% read, and even then only in very specific conditions.

[schumaku]

100.000 iops... yeah sure... with 8 SSD's in raid-0, 100% read, and even then only in very specific conditions.

I'm just citing marketing information ... and expect there is some backing available to proof

[ncvbit]

Thank you for the responses. I thought the 100.00 seemed a little high

To get the most out of a 2 qnap with 2 esxi host configuration would I spit the vmhds between the two datastores then? or is there a way to tell vcenter to divide the two evenly, datastore clustering?

Thanks,

Brandon

[murdocmk]

You'll have to divide the VMs between datastores yourself. If you have Enterprise Plus licensing with vSphere 5 you can use Storage Distributed Resource Scheduler, which will automatically shuffle VMDKs between datastores. But most customers will just be assigning VMs/VMDKs to datastores and moving them manually if they find them imbalanced in terms of space or IO demand.

Keep in mind that *most* VM environments result in a random IO pattern on your datastore. Even once you take a couple of sequential-pattern VMs and put them on datastore with other VMs you end up generally getting a random IO pattern. Have you done some measuring of IO metrics on your systems to plan your storage appropriately? Nothing is really a substitute for looking at your current patterns/load and then making sure you have the right disks/controllers/connections to get what you need.

Mark

[murdocmk]

As annoying as that was, that was pretty funny to read aloud.

Edit: This was in reference to some SPAM that is now missing. So that's why it looks out of place.

[ncvbit]

I ran a dell DPAC on our systems to generate our metrics. I am only consolidating 9 servers and couldn't justify an equallogic array. I paired my 879us with intel x520t2 10Gb nics. The only server that gave me trouble was exchange, which I reconfigured with more RAM and I split logs and DB on each qnap. Works great so far. All for 1/4 of dells suggested price. Once i have everything setup properly I was planning on doing up a writeup for anybody on the qnap bubble.

[murdocmk]

Great solution. Definitely interested in your write-up Ncvbit. Sounds like information the community could really benefit from.