pbulk on slower architectures

[John Klos]

Hi,

Is there a good / best practice way to do an initialization of pbulk on a
fast machine, then move that work to a slower machine?

An older machine, which is fast for a slower architecture (60 MHz m68060)
was started nearly a week ago, but has only gotten through 1600 out of
12217 packages...

Thanks,
John

[Aleksej Saushev]

Hello,

Post by John Klos
Is there a good / best practice way to do an initialization of
pbulk on a fast machine, then move that work to a slower
machine?

Pack the whole prefix with pbulk tools up and unpack it on target machine?

Post by John Klos
An older machine, which is fast for a slower architecture (60
MHz m68060) was started nearly a week ago, but has only gotten
through 1600 out of 12217 packages...

Well... I'd start from selected set packages on that...

[David Holland]

Post by John Klos
Is there a good / best practice way to do an initialization of pbulk on a
fast machine, then move that work to a slower machine?
An older machine, which is fast for a slower architecture (60 MHz m68060)
was started nearly a week ago, but has only gotten through 1600 out of
12217 packages...

I don't think you can, one of the reasons being that the results may
(will, in general) change when you switch architectures.

Ultimately we really need to make bmake run faster. Unfortunately,
it's a heck of a lot of work...

(the alternative I guess is to move away from bmake as the
implementation language, but that is also problematic and seems like
it would be a pity)

[Manuel Bouyer]

Post by David Holland

Post by John Klos
Is there a good / best practice way to do an initialization of pbulk on a
fast machine, then move that work to a slower machine?
An older machine, which is fast for a slower architecture (60 MHz m68060)
was started nearly a week ago, but has only gotten through 1600 out of
12217 packages...

I don't think you can, one of the reasons being that the results may
(will, in general) change when you switch architectures.
Ultimately we really need to make bmake run faster. Unfortunately,
it's a heck of a lot of work...

FWIW, I noticed that pbulk scan became much slower beteen netbsd-5 and -6.
I don't have statistics, but watching the bulk runs for official
builds is enough to notice it (the scan phase is reliably much
faster on -5 than on -6, and I noticed this on i386, amd64 and sparc).