Kernel 2.4: Load average scaling to 200!

Kernel 2.4: Load average scaling to 200!

[James Vanns]

All, 

As I am not sure whether this problem lies with the kernel core (process
scheduler, context switching etc. and all that other stuff that flies
over my head!), a scsi driver or the scsi tape driver or the 'mt'
command itself I have copied the kernel mailing list in this E-mail.

When using mt on RedHat Linux 8 running kernel 2.4.20 to fsf on an LTO
tape drive the load average soars to 100+ and denies everyone the mail
service it's used for (IMAP). I have to reboot the server as I cannot
kill mt. 

Has anyone had this kind of problem before? Any ideas if this is kernel
related (sorry to waste time if you are sure it's not)? I have looked
into mt's history to find no such bug report before.

Regards

James Vanns

-- 
James Vanns BSc (Hons) MCP
Linux Systems Administrator
Senior Software Engineer (Linux / C & C++)
Canterbury Christ Church University College
Public Key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x24045370

Re: Kernel 2.4: Load average scaling to 200!

[Alan Cox]

On Mer, 2003-09-03 at 17:17, James Vanns wrote:
> When using mt on RedHat Linux 8 running kernel 2.4.20 to fsf on an LTO
> tape drive the load average soars to 100+ and denies everyone the mail
> service it's used for (IMAP). I have to reboot the server as I cannot
> kill mt. 

I can think of two possible causes here. #1 is that something hung the
scsi bus because of a crash - be it kernel, mt, hardware whatever. The
other one some people hit with tape drives is that they don't have
disconnect enabled on the drive (check your drive manuals). With
disconnect disabled the tape will hog the bus for the entire time it
is doing a command.  

The tapes I have don't seem to take long enough to fsf (even if they are
past the last marker) to worry too much however.

Do you also get messages about scsi timeouts

What card ?