Bonnie_Amix2.1c_A3000D

Overall, performance is...really poor. The NFS tests took a really long time. I won't have a good idea just how bad until I get the other tests completed, but I recall various quips and quotes from Usenet and reviews indicating disk performance being slow. There are patches to improve performance, but I think these are for 2.01/2.03. If it's possible to apply them to 2.1c I'll provide results for that as well. On to the results.

Bonnie testing parameters

Bonnie was run with a 100MB test set. For AMIX it was compiled with gcc-2.7.2.3, using CFLAGS=-O2 -m68020 -fomit-frame-pointer.

Key to X axis labels:

OChr: Output a character at a time as fast as possible.
OBlk: Output intelligently as fast as possible.
RW: "Rewrite test" - Read some data, change it, write it, and read it again, intelligently, as fast as possible.
IChr: Read a character at a time as fast as possible.
IBlk: Read intelligently as fast as possible.


Bonnie performance on disk

First there's the disk performance using the UFS filesystem. I'm not going to bother testing S5 as it has so many limitations. Plotted are the rates in KB/sec and the average CPU utilization during the test.
Image

Bonnie performance on NFS

This was a test of NFS performance. The NFS server was an i686 Debian GNU/Linux 5.0 system with 100Mb NIC writing to a RAID 5. I assure you the bottleneck was not the Linux machine. Plotted are the rates in KB/sec and the average CPU utilization during the test.
Image

Performance summary, disk and NFS

This compares the data rates for disk and NFS, and disregards the CPU utilization. Plotted are the rates in KB/sec and the average CPU utilization during the test.
Image

Summary

I/O on Amix is relatively slow and uses relatively high amounts of CPU power to do the work. NFS is much slower. An advantage of NFS, though, is you can use RAID-backed storage at the other end. There is no such option for disk-based storage on AMIX.

Results for NFS were much tighter across 4 runs than the disk data. The first disk test for IBlk was a bit of an aberration, using far less CPU and having less speed. Probably another process was kicked off by the system during this test and multitasking caused Bonnie to get less CPU time.

Contributors to this page: failure .
Page last modified on Thursday 04 of June, 2009 19:58:31 CEST by failure. (Version 5)

History4 files attached (Hide)
List of attached files
 id Name   desc uploaded Size Downloads
1 11 default amix_bonnie_data_nfs_cpu View Download   Raw cpu util data Thu 04 of June, 2009 19:26 CEST by failure 100 b 292
2 10 default amix_bonnie_data_nfs_rate View Download   Raw NFS rate data Thu 04 of June, 2009 19:25 CEST by failure 104 b 265
3 9 default amix_bonnie_data_cpu View Download   Raw cpu util data Thu 04 of June, 2009 19:25 CEST by failure 100 b 253
4 8 default amix_bonnie_data_rate View Download   Raw disk rate data Thu 04 of June, 2009 19:25 CEST by failure 103 b 252