Rob van der Heij <[EMAIL PROTECTED]>  wrote :-

> Use the "random" stage to come up with random numbers, and prefix your
> data records with them and sort:
>   \ var list | split | s: spec stop anyeof select 1 1.4 c2d 1.10 r
> select 0 1-* n | sort | substr 11-*
>   \ random 100 | s:

> I hope this is not distributing over virtual machines, because then
> you should consider not keep all servers moderately busy but run just
> as many as you need (to let the rest sleep).

Rob,

Thanks, that is exactly what I wanted and, when I see it, it seems so
obvious.

Sorry, I did not explain the reasons for wanting to distribute the work
between the servers - which has nothing to do with workload.  These
servers run on a separate system, on physically separate hardware that has
enough resource.  The problem is that the work is sent via a UDP command
which then waits for a response or time-out. The time-out has to be long
enough for the longest task but, if the target server is not responding at
all for some reason, we get no error message but have to wait for the
time-out before trying the next. By spreading the sequence of the ports
(servers) we should get fewer such conflicts.

These are, in fact, remote compiler servers because we only run compilers
on a small z890 server and the remote compiles happen transparently using
replaced commands for such things as REXXC or HLASM.


Colin Allinson

Amadeus Data Processing GmbH

Reply via email to