[Swift-devel] Re: replication vs site score

[Mihael Hategan]

On Thu, 2009-04-09 at 09:52 -0700, Ioan Raicu wrote:
> What we usually do on the BG/P when using static provisioning, is that
> Swift does not start until the resources have been allocated, and that
> is because Falkon does not start until the resources are allocated.
> The whole process is automated, in terms of Swift waiting for Falkon
> to start, and Falkon waiting to start after resources get allocated.
> So, at Swift startup, in a static provisioned case, Swift could have
> all the information it might need, such as number of processors,
> number of sites, load (i.e. idle, as the resources are dedicated to
> Swift), etc.

You seem to be describing the scenario of not submitting a job unless
you know it will be executed immediately because there is an active
worker for it. Which I agree with.

Why I don't get is (and this is what I understand by "static
provisioning") where is the benefit in having a barrier that waits for
all requested workers to start, given that some workers will start
before others and will invariably have to sit idle until all workers are
started.