[OmniOS-discuss] ZPOOL disk performance on Supermicro with LSI 9300-8i

[Josh Barton]

We just built a Supermicro server with 3 LSI 9300-8i HBAs and 12 2 TB hard drives and are concerned about slowness after running a postgres benchmark. We also ran this benchmark on an HP Proliant 380 G9 with 14 disks and on a Sunfire server running OmniOS as well. The results showed the HP Proliant being many times faster (orders of magnitude) than the newer Supermicro server and the Sunfire.

Any ideas why the HP is so much faster? It just has one Smart Array Controller which I didn't think would be faster than JBOD

Results:
Sunfire
-bash-4.2# ./pgbench -c 50 -s 100 -t 100 -U postgres climate
Scale option ignored, using pgbench_branches table count = 1
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
query mode: simple
number of clients: 50
number of threads: 1
number of transactions per client: 100
number of transactions actually processed: 5000/5000
latency average: 0.000 ms
tps = 76.249498 (including connections establishing)
tps = 76.940244 (excluding connections establishing)


Supermicro Server
-bash-4.3$ ./pgbench -c 50 -s 100 -t 100 -U postgres climate
Scale option ignored, using pgbench_branches table count = 1
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
query mode: simple
number of clients: 50
number of threads: 1
number of transactions per client: 100
number of transactions actually processed: 5000/5000
latency average: 0.000 ms
tps = 119.929719 (including connections establishing)
tps = 120.786598 (excluding connections establishing)
tps is the metric we've been comparing (tx per second)

HP Server
root at cldb3:/usr/local/src/postgresql-9.4.4/contrib/pgbench#<mailto:root at cldb3:/usr/local/src/postgresql-9.4.4/contrib/pgbench#> ./pgbench -c 50 -s 100 -t 100 -U postgres climate
Scale option ignored, using pgbench_branches table count = 1
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
query mode: simple
number of clients: 50
number of threads: 1
number of transactions per client: 100
number of transactions actually processed: 5000/5000
latency average: 0.000 ms
tps = 1068.027362 (including connections establishing)
tps = 1254.114121 (excluding connections establishing)


Thank you all,

Josh
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://omniosce.org/ml-archive/attachments/20160622/cecef42d/attachment.html>