Unsolved
This post is more than 5 years old
1 Rookie
•
51 Posts
0
15167
February 28th, 2012 07:00
vmfs Benchmark changing over time
I'm in the middle of migrating esx4 to 5 and did some benchmarks again.
With SANHQ its nice to see whats going on on the san side.
All benchmarks shows as expected without the reallife benchmark with vmdk files.
I did the test on a new esx5 server with 2 san groups with one member each (ps3700 and ps4000) with MEM enabled and 4 gbit uplinks.
I had the same issue on another host with different hardware to the same san groups (also esx5 and MEM + 4 uplinks).
The first run on both was weird. At first it was slow with a queue depth of 1, but in the time it was going up. A second run just after the first showed better results.
I did not had this with iscsi volumes inside the guest vm.
1st run:
2nd run:
Anyone have a explanation?
Its no big deal, because the other benchmarks all showed fine...
some other screenshots:
100% read to 3700X 16x 400GB 10k SAS Raid 50 (with jumboframes):
100% read to 4000 16x 1TB 7200 SATA Raid 5 (without jumboframes):


sbuerger
1 Rookie
•
51 Posts
0
February 28th, 2012 07:00
1st run
2nd run
PS3700
PS4000
sbuerger
1 Rookie
•
51 Posts
0
February 28th, 2012 07:00
Somehow all screenshots were gone...
1st run:
sbuerger
1 Rookie
•
51 Posts
0
February 28th, 2012 07:00
Forgot the link to the benchmark: http://vmktree.org/iometer/
sbuerger
1 Rookie
•
51 Posts
0
March 1st, 2012 00:00
Looks like its the way the vmdk is created.
With lazy zeroed I get this slow first run with queue length 1.
With eager zeroed its okay at the first time.