Process Accounting on a Linux Cluster

[Gary Stiehr]

Hi,

1.) Has anyone had experience with process accounting on a Linux
cluster?  We are interested in knowing who is using what resources
(simply for our information).  I found a package psacct that is(was?)
maintained by the FSF. (I looked at gnu.org but didn't see much
information about it.)  Anyway, I can use this program on each
individual node and then collect all of the statistics individually, but
I was just wondering if anyone had come up with a different way (or if
you know of any other Linux process accounting packages). 

2.) Does anyone know how packages such as Mosix and Scyld Beowulf, which
modify the Linux kernel to handle processes, handle process accounting
when using packages such as psacct.  

3.) Does anyone limit the amount of resources that each user can use? 
Or assign certain nodes to certain users?  If so, which types of
programs do you use.

I appreciate any help anyone can give.

Thanks,
Gary Stiehr
Instructional Technology Services
University of Missouri - St. Louis
gary at umsl.edu