Ted Dunning a écrit :
Brice,
Looks like a nice piece of work. I just spent 20 minutes looking back for
the old bug HADOOP-2573 only to find that you already knew about that and
had addressed it.
This could really help people make progress on improving the scheduler and
that progress would really improve the usability of large clusters with jobs
that vary a lot in importance and size.
And your English is just fine. Completely understandable.
Thanks a lot for this very kind mail *^__^*
With Jaideep we are discussing of a way to describe ressources used by
Jobs (Network, LocalStorage, CPU and so on) in order to run more
dissimilar tasks (e.g. data intensive and cpu intensive) on the same
node. We hope to have a concrete proposition by Tuesday. When it will be
ready, I would really like your feedback of whether it would match your
needs or not, and if not what kind of schedulers you would need.
Regards,
Brice