"nice" I/O?

Clark Thomborson cthombor at luke.d.umn.edu
Sat Dec 23 07:33:42 AEST 1989


Problem: one I/O-bound process on one workstation can nearly saturate our
fileserver.  Two I/O-bound processes make life miserable.  I don't know of
any concept of "I/O priority" in either Unix or NFS that would limit the
frequency of fileservice for such processes.  Is there such a notion
anywhere in the Unix world?

Solution(?): I can imagine a new "limit" parameter, pagefault_frequency.
A process with a pagefault_frequency of 5, for example, would get no more
than 5 pagefault requests each second from its local Unix kernel...the
sixth would be delayed until the next second.

Can anyone think of a simple workaround that doesn't involve kernel
hacking?  I don't want to wait years for a solution.



More information about the Comp.sys.sun mailing list