Check out the new USENIX Web site.
... Alicherry1
The author is currently with Bell labs, Murray Hill, NJ. The author can be reached at mansoor@research.bell-labs.com
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... Gopinath2
The author can be reached at gopi@csa.iisc.ernet.in
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... tasks3
Soft real-time tasks do not have ``strict'' deadline requirements, though some timeliness requirements still exist.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... time.4
We collect the resource usage of resource containers for CPU time, network bandwidth and disk usage. This can be used for computing the priority of the resource container if appropriate weights are assigned to each of the resource usage. Also the network bandwidth usage will reflect on the CPU usage since the protocol processing for the packets are significant and we account for the network processing properly.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... containers5
The descriptions in this section closely follow those in [1] & [6].
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...6
Eventhough we use CPU time stamp counter for keeping track of time, we get timer interrupts only in 10ms intervals.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... client7
This application recognizes digits using a number of computational nodes (clients) with the server using many clients. Most of the time server sleeps. We have run all the clients on the same machine.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.