Measuring open/close IO performance

Local/NFS performance measured 2009-02-06

[cycles/s] h1 h2Sorted ascending h4 remarks
s01 108 3.5 101 via NFS, zpool setup with 3 large vdevs
s13 109 4.1 115 via NFS, zpool setup with 8 small vdevs
d02 832 4.5 631 via NFS, Areca 16 disk RAID6
local 59282 3586 60301 h1,h2: Areca 4 disk RAID6, h4: Areca 8 disk RAID6
KERNEL 2.6.24.4 2.6.27.7 2.6.27.7 kernel at the time of testing

Measurements are done with this stupid little program

One word of caution concerning h2: This machine was always under a high load of 20+ (sometimes briefly up to 100) while tests were run. Big question is why.

local IO performance on h4

On the local h4 RAID6 with 8 disks the open/close program was run several times along with stress with either the --cpu x or the --vm x options. As soon as the number x comes close to the number of system cores the performance drops down significantly:

openclose.png

-- CarstenAulbert - 06 Feb 2009

Topic attachments
I Attachment Action Size Date Who Comment
measure_open_close.cc measure_open_close.c manage 964 bytes 06 Feb 2009 - 15:32 CarstenAulbert Little program opening and closing the same file over and over again
openclose.pngpng openclose.png manage 29 K 06 Feb 2009 - 11:09 CarstenAulbert Open/close performance versus stress test utility
This topic: ATLAS > OpenCloseIotest
Topic revision: 16 Feb 2009, CarstenAulbert
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback