|
| Hi there,
I'm a computational chemist and I've created a small cluster. I do
serial batching jobs, e.g. I distribute Gaussian03 jobs to single
computing nodes, were they run on 1 processor; very simple batching
system. The scheduler I use is a very simple one, which checks
5mloadavg for all the resources every 5 minutes (probe).
I'm writing my PhD thesis at the moment (chemistry topic btw) and I'm
also writing something about my cluster creation. I've read a lot
about metrics and performance for parallel computation (Amdahl etc.).
Based on the same principle I devised some metrics for my own
scheduler, i.e. an expression for the maximal and minimal speedup in
function of probe time and job length.
Now am I looking for references to metrics/performance descriptions
(not benchmarks) for batching systems in general, because I have a dark
blue feeling that I'm inventing the hot water again ... (or maybe cold
water).
I did a massive amount of reading on parallel and distributed systems
(being a chemist) but I can not seem to find any references in
literature to batching systems metrics.
I hope I made myself understandable (not controlling the jargon 100%).
Thanks in advance,
TK
--
http://users.UGent.be/~tkuppens/
---
Save our planet, it's the only one with chocolate!
-- Anonymous
|
|