OI/Napp-it on recycled hardware pwns EQL SAN

jb33

n00b
Joined
Feb 6, 2013
Messages
39
IO Meter test is run from a ESXi 5.1 guest w an OI hosted iSCSI target. OI server specs are as follows: 6 2TB 7.2K Discs / 48gb RAM, 2x 2.33 Ghz Quad-core processor, 1x1Gb iSCSI NIC.

I think when I figure out how to multipath a second NIC the OI box may dominate conclusively. But the iops and latency numbers on the OI tests seems almost inverted to me. Anyone have any insights?

OI RAID Z2
Test name Latency Avg iops Avg MBps cpu load
Max Throughput-100%Read 17.16 3481 108 13%
RealLife-60%Rand-65%Read 4.75 12240 95 47%
Max Throughput-50%Read 12.13 4946 154 13%
Random-8k-70%Read 4.84 12024 93 44%

OI RAID 10
Test name Latency Avg iops Avg MBps cpu load
Max Throughput-100%Read 17.25 3461 108 15%
RealLife-60%Rand-65%Read 4.47 13036 101 50%
Max Throughput-50%Read 11.95 5012 156 12%
Random-8k-70%Read 4.56 12798 99 50%

Equallogic PS6100E / 24 1TB 7.2K discs / RAID 10 2x1Gb iSCSI NICs
Test name Latency Avg iops Avg MBps cpu load
Max Throughput-100%Read 8.53 6704 209 10%
RealLife-60%Rand-65%Read 12.73 3535 27 18%
Max Throughput-50%Read 8.70 6606 206 29%
Random-8k-70%Read 12.91 3543 27 17%
 
latency and iops seems ok for zfs system, you are probably hitting some read requests from ARC cache.

you need to know size of your working set, ratio for random/seq and ratio for read/write to directly compare what will be better for you in production.
 
Back
Top