Messages in this thread | | | Date | Fri, 16 Jul 1999 17:41:49 -0400 (EDT) | From | "Richard B. Johnson" <> | Subject | Re: OOM |
| |
On Fri, 16 Jul 1999, Bernd Paysan wrote:
> On Fri, 16 Jul 1999, Claus Fischer wrote: > [SNIPPED]
> > This is just a sort of conservative capitalistic/aristocratic algorithm. > If all cheap resources are occupied, allocating new resources becomes more > expensive. The idea is to keep the setting in a sane state by killing > newcomers. People don't want to get their two-week simulation killed, > therefore Rik's algorithm takes total cpu usage (processes with more > computing power invested are worth more) and runtime into account. >
I don't like the idea of killing off tasks that request resources that are not presently available. I think they should sleep (interruptable) until resouces are available -if ever. This allows the user to determine if he/she wants a particular task to continue.
This means that a malicious hog will just go to sleep. This was one of the nice things about VAX/VMS. Just swap the whole thing to 'mechanical memory' until resources became available. Under VMS, this was handled as SRPs and LRPs (small and large request packets). Small packets were practically always available (512 bytes on VAXen). Large packets might not be and could force a, sometimes very long, wait.
When I was at GE's Advanced Development Group, working on Ultrasound simulations, I had a FORTRAN program that would take about a week to complete. When it completed, often a few terminals in which persons had attempted to log-in over the past week (and gave up), would finally come up with the '$' prompt as resources became available to create the new processes.
Cheers, Dick Johnson FILE SYSTEM MODIFIED Penguin : Linux version 2.2.6 on an i686 machine (400.59 BogoMips). Warning : It's hard to remain at the trailing edge of technology.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |