[Swift-devel] pilot job paper

[Mihael Hategan]

On Wed, 2011-07-20 at 16:41 -0500, Ketan Maheshwari wrote:

>         
>         The number of jobs is an orthogonal dimension (i.e. we may
>         want to
>         measure the queuing time vs. #of jobs for various walltimes,
>         but later).
>         The actual job duration is not relevant. The amount of data is
>         not
>         relevant.
>         
>         Clouds are an interesting environment, but not for this
>         particular
>         problem. That's because we need to see how much it takes to
>         acquire
>         resources, not how fast some job middleware is after we got
>         hold of
>         those resources.
>         
> 
> Do you have any specific environments in mind for these experiments?
> For the requirement of various/random degrees of utilization, we can
> use MCS local cluster (10 x 64bit + 3 x 32 bit machines), Beagle, and
> Ranger. 

Not really besides them being clusters with multiple users.

When there are fewer resources than demand, some economics needs to
happen. This is mostly visible in queuing times (a user has nothing else
to pay with than time - assuming a working environment). There are
multiple things that influence that cost, but I only want to know how
requested walltimes affect it. Since we can't directly isolate that from
other things, he have to employ statistics. The more random the
clusters/times/<any other variables that could affect queueing times>,
the better we can infer some general formula for things we're interested
in (because whatever is common to all these environments is likely to be
the effect of the parameter we're trying to play with - walltime).