Latest interface: 0.3.1
Latest system: 010
fluppie007
User

52 posts

Posted on 21 January 2014 @ 04:13
ZFSguru 0.2.0-beta9 (9.2-001) pool benchmark
Pool : storagepool (16.2T, 79% full)
Test size : 256 GiB
normal read : 361 MB/s
normal write : 249 MB/s
I/O bandwidth : 31 GB/s

RAID-Z2 pool with 6x 3TB disks (ATA Hitachi HDS5C303 AAB0) virtualized under ESXi 5.0 Update 2.
The VM has 16GB RAM and 2 Vcores. The 6 disks are connected to a IBM M1015 which is passed through using VT-D.

Is this acceptable performance or should I cry :D?
karmantyu
User

131 posts

Posted on 21 January 2014 @ 10:56
I dunno, for comparison I made a test on a real system for you.
ZFSguru 0.2.0-beta9 (9.1-005) pool benchmark
Pool : TANK (10.9T, 66% full)
Test size : 256 GiB
normal read : 295 MB/s
normal write : 153 MB/s
I/O bandwidth : 2 GB/s

on a real HPn40L RAID-Z2 pool 6x2TB WDGreen system booted from USB
CiPHER
Developer

1199 posts

Posted on 21 January 2014 @ 14:17
Both good scores; fluppie's pool is more full and virtualized under ESXi. So it seems that 250MB/s write for a 256GiB huge test file is quite good. On shorter writes the score will be higher, but this is true sustained performance over long time.

Karmantyo has less powerful system and probably fewer memory as well. Look at the I/O bandwidth: the hardware of fluppie is more than 10 times faster. So this will limit the write score particularly, due to the heavier calculations. But nonetheless reasonable score for a low-power server.
karmantyu
User

131 posts

Posted on 22 January 2014 @ 11:29
:) "less powerful" nice. It was about $150 cheap, and it has very low power consumption, but stuffed with 16GB ECC RAM too. Off course it is slow like hell, but good enough for home.
CiPHER
Developer

1199 posts

Posted on 22 January 2014 @ 12:07
For gigabit that is just great indeed. By the way, i managed to get 8,4 watt on a Intel Ivy Bridge quadcore (3570) using BIOS undervolting, no hardware mods. Using PicoPSU power supply and basically everything turned off in BIOS. Still a nice score not many people can beat.

So it is possible to combine low idle power with powerful components, but it is tricky. The power supply alone can mean double the power consumption at these low power states.
Last Page

Valid XHTML 1.1