Subject: Re: Fixing the swapper (was Re: Summarisation on uvII/VS3100
To: Jason Thorpe <email@example.com>
From: Chuck McManis <firstname.lastname@example.org>
Date: 06/23/1999 09:28:58
At 07:59 AM 6/23/99 -0700, Jason Thorpe wrote:
> (1) What *exactly* is the symptom you are observing, and
The exact symptom is that a process goes to sleep waiting for a buffer
(biowait) and never awakens.
> (2) What makes you think the "swapper" is the problem?
If you don't swap you don't fail.
Now these symptoms have been observed (and I can reproduce them) on both
the KA630 and KA650 CPUs. I've not tried on my KA655 but it is similar
enough to the 650 that I don't think there will be much difference.
There is another, possibly unrelated, problem with my MSCP disks behind a
Sigma controller that has been duplicated in at least one other site (also
with a Sigma controller) The symptom is that processes get stuck in
diskwait, (often on buffers that are marked for delayed write) and the MSCP
driver never writes them and frees them. This is the problem I am currently
investigating, although the lack of a public MSCP spec hinders that
developement (I re-iterate my call for anyone with the internal DEC spec,
the patent runs out this year I believe and DEC doesn't sell MSCP based
devices as far as I can tell, so please let us see it.)
Anyway, my current attack plan is to substitute a "real" DEC MSCP based
controller and to verify the problem is either fixed or goes away. If it is
fixed I'll use an MSCP traffic inspector in the driver to watch the packets
and the results and compare them to the Sigma board.