Storagespaces: CRC Errors on initiator

[qwertz]

Hi there!
Thanks for your reply.
As my other lab is currently occupied with some test systems running for my academic studies, I'm setting up a "new" lab (again with tiered storage spaces) to test that behavior again.
Student -> Limited Budget -> old hardware.

I've got 2 Supermicro X7DB8 Boards, 4x Xeon E5420, 2x 8 GB DDR2-ECC RAM and 2x Brocade 1020 to play with.
I also got enough hdds to build 2 storage spaces, but in order to build the same storage space again... I also need 2 SSDs.
And I need to save more money to buy them... Maybe in 2-4 weeks...

Thats how it looks like right now:


I'll keep you updated as soon as I got this lab together.

[anton (staff)]

Reminds me my old good days in college

[qwertz]

Setup has progressed to the state of waiting for the ssd's.
I've tested (Memtest, prime) the hardware and installed 2 win2k12r2 servers.
thats how it looks like now:

[anton (staff)]

When SSDs would arrive it would be nice to see some raw benchmarks before building a hypervisor-converged config

[qwertz]

Hi there!
Finally They, 2x Samsung Evo 840 SSD - 250GB, have arrived.
And, I did some initial benchmarks.
Here are the Results from Atto:

atto_bench

840_ntfs_compared.png (31.78 KiB) Viewed 22347 times

They are not as fast as they could be: only got s-ata II on those boards.
I also noticed that those CPUs aren't exactly the same. left one has 2x E5405 right one has 2x E5420.
Looking forward to do some basic iops testing with iometer.
But for now, iometer crashes if I try to perform write activity.
(I was hoping to test 80% write 20% read, 80% random 20% sequential, 4k transfer size and boundaries 4k...)
Didn't figure out why so far.
However if I perform read only testing with iometer the values differ VERY much.
(I'm testing on ntfs volumes, not raw!)
Have a look:

iometer_results_compared

840_iops_compared_READ.png (44.83 KiB) Viewed 22347 times

As soon as I find more time, I'll dig a little more into this. Any other recommendations how to measure iops?
Something else I should test?
Kind regards!

[anton (staff)]

I'd use SQLIO as it's what MSFT loves to show on their demos (you have a great chance to publish whole set of information on your blog BTW, we can even $$$ sponsor this event).

Run tests in a loopback. Run tests over Ethernet with a FLAT and LSFS containers.

We've found some problem with kernel performance on StarWind so you're not going to squeeze more then 100K IOPS from a single LU so far with a flat non-cached content