2002-09-05 11:46:10 +00:00
|
|
|
<title>Scheduling server: policy</title>
|
2002-08-20 23:54:17 +00:00
|
|
|
<body bgcolor=ffffff>
|
2002-09-05 11:46:10 +00:00
|
|
|
<h2>Scheduling server: policy</h2>
|
2002-04-30 22:22:54 +00:00
|
|
|
<p>
|
2002-09-03 21:27:21 +00:00
|
|
|
The scheduling server will attempt to send enough work to exceed a
|
|
|
|
|
hosts high water mark. If the amount of work the scheduling server is
|
|
|
|
|
sending exceeds a certain level (default is four weeks), the
|
|
|
|
|
scheduling server will not attach more work to a scheduler reply.
|
|
|
|
|
This does not prevent the scheduling server from sending a lengthy
|
|
|
|
|
workunit, but rather from sending multiple lengthy workunits.
|
|
|
|
|
<p>
|
2002-07-29 19:01:38 +00:00
|
|
|
If a work unit uses more disk resources than a host has available, the
|
2002-08-19 18:43:10 +00:00
|
|
|
scheduling server will not attach that work unit.
|
2002-09-03 21:27:21 +00:00
|
|
|
<p>
|
2002-08-19 18:43:10 +00:00
|
|
|
The scheduling server
|
2002-07-29 19:01:38 +00:00
|
|
|
estimates the amount of time a work unit will take to complete with the
|
2002-09-03 21:27:21 +00:00
|
|
|
formula <b>(number of flops)/(flops per second)+(number of iops)/(iops
|
2002-08-19 18:43:10 +00:00
|
|
|
per second)</b>.
|
2002-09-03 21:27:21 +00:00
|
|
|
The number of floating point and integer operations are provided by
|
|
|
|
|
the project when creating the work unit, and the host calculation
|
|
|
|
|
speeds are included in a scheduler request.
|
|
|
|
|
<p>
|
|
|
|
|
If no work is available, the scheduling server sends the message
|
|
|
|
|
<tt>no work available</tt> with priority <tt>low</tt>, and
|
|
|
|
|
requests that the client wait before sending another request.
|
|
|
|
|
<p>
|
|
|
|
|
If there is work available, but the host cannot accept it for whatever
|
|
|
|
|
reason (too slow, not enough space, etc), then the scheduling server
|
|
|
|
|
sends the message <tt>host unacceptable</tt> with priority <tt>low</tt>,
|
|
|
|
|
and requests that the client wait before sending another request.
|
