Jay.Estabrook@digital.com
Thu, 15 Apr 1999 13:15:03 -0400
>>> Greg Lindahl said:
> > > * Building a kernel from clean is about 2-3 minutes with -j 8. Much
> > > more than 8 and gmake looses it, emitting nonsense errors, etc.
> >
> > I also would like to understand why the 2.2 kernels changed the way parallel
> > makes work, as the 2.0 kernels are very happy with just -j. Anyone know the
> > history behind this?
>
> Um, is this still broken after your patch? I would think that high
> load would bring out the bug you zapped.
No, no, that's not what I meant...
The 2.0 kernels did "parallel make" via breadth-then-depth; when dropping
into a subdirectory, compile jobs would be spawned for all the files in that
directory, and then wait for those to finish before going on to
subdirectories or popping back up to do any siblings, etc, etc. So
essentially the number of parallel compiles was limited to how many compiles
were at a certain directory level, ie drivers/scsi would spawn off compiles
for all the drivers configured plus the "generic" SCSI subsystem sources
(like scsi.c and hosts.c). This never got so big to cause problems; I like
to run a pair of parallel makes in infinite loops to put a test load on a
system.
Now, the 2.2 kernels changed to a depth-then-breadth technique (at least
that's what it looks like to me), in which all directory levels are
activated at the same time (unless limited by a max number of processes to
spawn, ie -j 8), with one compile at each level. This results, with just -j
and no max limit, in a veritable flood of processes/compiles starting up,
and it seems to overwhelm some part of the system.
So, I ask again, anyone know why this was done, and does it adversely affect
the INTELs as much as it does the ALPHAs, if just done with -j?
--Jay++
-----------------------------------------------------------------------------
American Non Sequitur Society: we don't make sense, but we do like pizza...
Jay A Estabrook Alpha Motherboards - LINUX Project
Compaq Computer Corp. (508) 841-3241 or (DTN) 237-3241
334 South Street, Shrewsbury, MA 01545 Jay.Estabrook@digital.com
-----------------------------------------------------------------------------
-- To unsubscribe: send e-mail to axp-list-request@redhat.com with 'unsubscribe' as the subject. Do not send it to axp-list@redhat.com
This archive was generated by hypermail 2.0b3 on Thu Apr 15 1999 - 11:00:07 PDT