DOCUMENTATION - v0.1
The documentation of the
classes generated by javadocs can be found [here].
Naturally, only this
documentation is not enough. So, below you can find some other useful
information on PaJFit. I will try to include more details according to
my time availability.
WORKLOAD FILE: This
file contains the list of jobs to be scheduled. Each line is a
different job
FOR RIGID:
subtime starttime priceprofile deadline budget numnodes reqtime
FOR MOLDABLE:
subtime starttime priceprofile deadline budget minnodes maxnodes MK MC
- subtime: submission time.
- startime: starting time. If startime is -1, it means that
the job is an advance reservation.
-
priceprofile: it is possible to specify different classes of
prices for jobs.
- deadline: maximum completion
time. If the
job is an advance reservation, deadline is ignored.
- budget: maximum amount of
money the user will spend to execute the job
- numnodes: number of required
nodes
- reqtime: amount of required
time to execute the job
For moldable jobs:
- minnodes and maxnodes: minimum and maximum
number of nodes
- MK and MC: parameters of function to
calculate the required time according to number of nodes:
Example: reqtime = MK - MC * numnodes
It is also
possible to specify the number of nodes in the system by using 'nnodes
xxx' where xxx is the number of nodes.
Example of
content with three jobs:
nnodes 64
1
2500 3.000000
43200 14400
5
5
2880 0
2
500 3.000000
43200 14400
10 2880
464
-1 1.000000
34400 19184
5
5
2880 0
SCHEDULING:
In order to fit a job, the scheduler looks for free nodes in each
scheduling event time. These
event times are places in the time axis where a job finishes the
execution. The first event time can
also be the current time or starting time when scheduling an advance
reservation.
Both on-line and batch job modes can be used. It is just a matter of
modifying the workload to have all the jobs with submission time equal
to zero.
Last update: April 11, 2007 by
Marco Netto