|T O P I C R E V I E W
||Posted - 04/05/2012 : 10:51:16 AM
I'm doing some disk benchmarks and I'm getting some weird figures that I could use help with.
Background: 2 almost identical physical servers (Dell R610) with 4x500GB SATA 7.2k disks in RAID on a integrated H200 RAID controller (without cache). Both have 2 quad-core Xeons, 48GB RAM (32 usable by the OS). Server1 is a production server with SQL+ISS workloads and currently has no performance problems. Server2 had SQL+ISS workloads running on it too, but customer complained about disk perf. All workloads have been moved from the server since then and it's idle now.
I've benched both with several tools - Passmark Peformance test 7.0 (64bit), IOmeter, ATTO HDD benchmark.
All benchmarks on server1 have been performed several times and the results is the average.
The results are:
Sequential read - Server1: 17MB/s, Server2: 175MB/s
Sequential write - Server1: 145MB/s, Server2: 50MB/s
Random seek + RW - Server1: 135MB/s, Server2: 10MB/s
Server2 figures seem plausible, on Server1 the bench seems to be broken (as stated above, it was run multiple times).
I've defined custom access specifications in order to mainly compare IOps.
100% Random 100% Read 4K - Server1: 470 iops, 1MB/s
100% Random 100% Read 4K - Server2: 520 iops, 2MB/s
100% Random 50% Read 4K - Server1: 330 iops, 1 MB/s
100% Random 50% Read 4K - Server2: 470 iops, 1 MB/s
100% Random 100% Write 4K - Server1: 254 iops, 1 MB/s
100% Random 100% Write 4K - Server2: 420 iops, 1 MB/s
100% Random 100% Read 32K - Server1: 320 iops, 10 MB/s
100% Random 100% Read 32K - Server2: 358 iops, 11 MB/s
100% Sequential 100% Read 4K - Server1: 27000 iops, 120 MB/s
100% Sequential 100% Read 4K - Server2: 11400 iops, 44 MB/s
100% Sequential 50% Read 4K - Server1: 52200 iops, 200 MB/s
100% Sequential 50% Read 4K - Server2: 1600 iops, 7 MB/s
100% Sequential 100% Write 4K - Server1: 67000 iops, 264 MB/s
100% Sequential 100% Write 4K - Server2: 2200 iops, 9 MB/s
For the 100% random tests, the results are virtually the same, the situation changes quite a lot with the 100% sequential tests. Server1's iops and MB/s seem ridiculously high and Server2's ridiculously low.
Finally, I've tested both server with ATTO (Server1 is on the left, Server2 on the right):
Both these look wrong to me too.
What I'm really trying to determine is if Server2's disk perf is adequate to the configuration (RAID10 from 4 SATA drives).
Any insight is appreciated.
|3 L A T E S T R E P L I E S (Newest First)
||Posted - 04/05/2012 : 2:34:25 PM
I know Evgenij, don't blame me though, I wasn't the one who bought it :)
||Posted - 04/05/2012 : 2:05:54 PM
Sorry for being harsh, but the H200 really is a piece of sh!t.
||Posted - 04/05/2012 : 11:52:17 AM
> 100% Sequential 100% Write 4K - Server2: 2200 iops, 9 MB/s
That's broken, especially when you get such different results between similar machines. You get better performance from the cheapest USB drive. Maybe get a real RAID controller? Or, did you maintain the firmware levels?