Re: Squid Cache Partition Performance Comparison of Reiserfs andExt2

Thanks for your response.

Yup, I figured I had two problems. The first being on the squid box itself with a disk bottleneck as this test was NOT designed for a mere IDE disk system. I discovered today with a "hdparm -t /dev/hde" on my squid box, that I was getting only 2.60MB/sec in that rough test, in comparison to 4.00MB/sec on the crappy little 200mhz workstations with old IDE controllers and drives. Definitely something was wrong.

I applied a flash update of the Abit BE6 BIOS and onboard HPT366 IDE controller, and this allowed me to change the disk mode on the 20.5GB IBM 7200rpm UltraATA/66 hard drive to something called MW DMA 2 (I have no idea what this is, anyone have any recommendations? Other options were DMA 0-4, PIO 0-4... etc). With the new IDE controller setting, I was able to achieve 9.60MB/sec with the same hdparm command. A great improvement, so I ran the tests again.

The system was already set to ext2, so I ran that test first. Remarkably there were virtually zero errors for most of the test. This was a GREAT improvement over the last ext2 test where it had MANY errors due to the overloaded disk bottleneck. When it hit 51 minutes of this one hour test it suddenly died. Looking at the screen on the server it displayed many disk access errors and the machine crashed. I restarted it to find the hard drive dead. It no longer spins up, and the BIOS on three computers that I've tested no longer recognize the drive.

In a way I'm glad it died now rather than later when this server would go into actual usage. These tests were meant to aid in performance tuning and stress test my rickety non-standard system. I'm guessing that this disk was a dud to die after only a few hours of heavy stress tests.

Anyway, with this new information I will try the testing again tomorrow when I get a new disk. This should result in some better results in comparing the actual filesystems.

Warren Togami
warren@togami.com

----- Original Message -----
From: "Jens-S. Voeckler" <voeckler@rvs.uni-hannover.de>
To: "Warren Togami" <laven@hawaii.rr.com>
Sent: Monday, December 27, 1999 12:12 AM
Subject: Re: Squid Cache Partition Performance Comparison of Reiserfs andExt2

On Sun, 26 Dec 1999, Warren Togami wrote:

]of these errors are the result of "Too many open files" errors on the
]client machine.

You need to correct the FD situation (see Henrik's page on details) on all
machines: clients, servers and cache. As long as the error rate in an
experiment is above 1 % (the number depends on what your directives are),
the results can be readily discarded. They are not comparable, and you
should not make any interpretations on data laced with errors. Rather, you
need to repeat the experiment until you get an error rate low enough.
Unfortunately, you need to repeat both experiments in the same
environment.

Le deagh dhùrachd,
Dipl.-Ing. Jens-S. Vöckler (voeckler@rvs.uni-hannover.de)
Institute for Computer Networks and Distributed Systems
University of Hanover, Germany; +49 511 762 4726
Received on Mon Dec 27 1999 - 17:13:11 MST