Data corruption on Tyan Thunder LE

G

Guest

Guest
Hi all-

I need independent verification of a data corruption problem with the Tyan Thunder LE mobo during DMA writes to an IDE drive. The problem manifests itself under heavy disk load in large files, does not change file size, and only corrupts 4 byte chunks. Thus, it is not easily spotted.

We are running redhat 7.1, but this corruption may be reproducible on other operating systems.

A simple way to verify this problem (in Linux anyway) follows:

Using a redhat 7.1 install (2.4.2 kernel, but other 2.4 kernels exhibit this problem) and a file /tmp/x that is over 500MB in size

#!/bin/csh
while 1
cp /tmp/x /tmp/y # copy the big file
ls -l # list the directory to show file sizes
md5sum /tmp/x # compute md5 sum for original file
md5sum /tmp/y # compute md5 sum for duplicate file
rm -f /tmp/y # remove dupe file and then do it all over
end

/tmp/x and /tmp/y will continuously show the same size, but will sometimes have different md5 checksums. A diff between the files usually shows that 4 bytes are moved to another part of the file. You may have to go through a number of iterations before you see it. We have a more sophisticated test script available, just e-mail me at wezelboy@cse.ucsc.edu and I'll send it to you.

If anyone can perform a similar test using other operating systems, I would appreciate it.

Tyan claims that this problem is a limitation of the Serverworks LE chipset, however I have not been able to reproduce this problem on an Asus CUR-DLS mobo (which uses the same chipset) If anyone can reproduce this problem on a different Serverworks LE mobo, I'd like to hear about it.

Thanks!

wezelboy
 
This is a link to a member of toms hardware community. He has a lab to run such tests.
http://stabletechnologies.com/
Also send your request to the webmaster at Toms Hardware, Hard OCP etc... If this is a true bug one of them should be able to reproduce it.
This type of request in other forums could cause terrible rumors to spread.


I aint signing nothing!!!