[cgl_discussion] Re: why strict_memory_overcommit patch?

Peter Badovinatz tabmowzo at us.ibm.com
Fri Sep 27 12:28:45 PDT 2002


Robert Love wrote:
> 
> On Fri, 2002-09-27 at 14:54, Randy.Dunlap wrote:
> 
> > Why do we have the strict VM overcommit patch in the CGL kernel?
> >
> > What requirement is it for?
> >
> > Or does some other requirement depend on it?

These are the sorts of edge conditions that we would very much like to
identify and call out as "interesting" issues.  But, no, no one thought
to bring up the specific point of what does Linux do when OOM and what
would enterprise/carrier servers desire to do in such a situation.

Call it a big miss in 1.1...

> 
> Back in the TLT days, someone at Intel filed a bug relating to OOM and
> someone else at Intel filed a reply saying "rml has a strict overcommit
> patch that should fix this" - so I merged it.
> 
> I would think its just a general robustness / stability sort of thing.
> The non-deterministic behavior of the OOM killer surely is not ideal for
> a telecom server.

Good point.  However, CGLE is not the end-all in running on telecom
servers.  Since we did not identify this behavior as something that we
want to describe we don't really have a connection to anything in the
Architecture/Requirements documents.

I don't think it buys us anything unless we can highlight that we've
dealt with this kind of condition, so it's being in CGLE isn't as useful
to us.  Having architecture/requirements coverage of it is a better
thing because it's really the specs and descriptions there that we want
followed.

> 
> But, no, I do not think it was a specific line item and nothing depends
> on it.
> 
>         Robert Love

Since we (OSDL CGL) have not identified and specified behavior of this
edge condition, IMO it's not our place to modify the base.  It is valid
for different distributors to do 'something' about it, as it's from them
that customers purchase their Linux version.  For now, it's a
differentiator they can claim, and as we've not described HOW they
should deal with it, the distro does that.

I think that we need as part of OSDL CGL Effort 2 a wide consideration
of "edge conditions" and how Linux should react so that vendors,
application writers etc. can expect deterministic behavior.  Or codify
that they'll not get deterministic behavior.


Peter
--
Peter R. Badovinatz aka 'Wombat' -- IBM Linux Technology Center
preferred: tabmowzo at us.ibm.com / alternate: wombat at us.ibm.com
These are my opinions and absolutely not official opinions of IBM, Corp.



More information about the cgl_discussion mailing list