INFO on the 4KB random project.
IOmeter configs by GullLars.
The point of this project is to chart 4KB random read scaling of various drives and arrays.
There are 2 series of configs, one for arrays off controllers with cache, and one for drives and arrays off controllers whitout cache or cache smaller than 256MB.
The cache series of configs use a 10GB testfile, 5 sec ramp, and 25 sec runtime for 10mins total divided by #workers.
The NOcache series of configs use a 1GB testfile, 1 sec ramp, and 9 sec runtime for 100sec total divided by #workers.
Access specification: 4KB, 100% random, 100% read, 4KB alligned IO's
Queue Depth (# of outstanding IO's) scales exp2 from 1*#workers to 1024. This is to allow mapping of the full scaling of larger arrays.
To make it possible to handle results from many people, the naming of result files will be as follows:
{Username}_{#of drives}{RAID level}_{drive name & size}_{RAIDcard/HBA name}_{stripe size}_{cache type}.csv
Example: GullLars_2R0_Mtron 7025 32GB_SB650_SS32KB_WT.csv
Include a .txt file with additional info on array, like drivers, firmware, read ahead, OC settings, etc.
To avoid large numbers of result files pr user for those running multiple configs (f.ex. 1/2/4/8 workers) for the same array, you can use the same result file for all runs on the same array as long as the settings are unchanged. This is done by using the same name for the result file. The new results are added at the bottom of the file and don't overwrite the old.
Reply With Quote
Originally Posted by GullLars
Bookmarks