Summary of Recent Iozone Results on Linux-2.6.0 Kernels

These tests were run through the Scalable Test Platform at OSDL.

About these Reports

Iozone is an IO test. It measures 'operations per second' where the 'operations' are the various types of reads and writes. These summaries report the percent difference for each kernel tested from the chosen baseline kernel. The 'Ave %' is the most useful value and is highlighted. It is the average percent difference from the baseline kernel. The Low % and the High % define the boundaries of the range of the percent differences averaged. Put another way, the 'low %' is the worst performing compared to the baseline and the 'high %' is the best performing compared to the baseline. All values are calculated from no fewer than 5 data points.

The reads and writes per second should preferably increase in the newer kernels; therefore for 'Ave %' bigger is better and negative shows regression. A value of zero means no change. For the baseline, the values will always be zero.

The reports have links to the original data and to information in the PLM about the kernels tested.

About the Testing

Files sizes used are from 1M to 16M for 4k records and 1M to 1G(2-CPU) or 512M (1-CPU) for 128k records. Information on the systems is available at http://www.osdl.org/lab_activities/kernel_testing/stp when you request a run.

The 2-CPU hosts are 850 MHz Pentium III with 256k cache. They have 1024M RAM. The disk we are using for the IO tests is an AMI MegaRAID controller which has 5 disks attached with RAID0 configuration.

The 1-CPU hosts are 1000 MHz Pentium III with 256k cache. They have 512M RAM. The disk we are using for IO tests has IDE 19G.

The 2-CPU systems are all done on a single host, and so are consistant. Results on the 1-CPU systems should only be compared for the same hosts. I am looking at trying to make these test results more reliable.