Path: utzoo!utgpu!utstat!jarvis.csri.toronto.edu!rutgers!deimos.cis.ksu.edu!uxc!uxc.cso.uiuc.edu!mcdurb!aglew
From: aglew@mcdurb.Urbana.Gould.COM
Newsgroups: comp.arch
Subject: Re: Programming (was Register usage
Message-ID: <28200328@mcdurb>
Date: 2 Jun 89 02:24:00 GMT
References: <10534@ihlpb.ATT.COM>
Lines: 16
Nf-ID: #R:ihlpb.ATT.COM:10534:mcdurb:28200328:000:879
Nf-From: mcdurb.Urbana.Gould.COM!aglew    Jun  1 21:24:00 1989


>From time to time, I have seen the argument that large register files slow
>down context switches.  Yes and no.  If you are looking at general purpose
>operating systems like Unix, register saves are a very small part of the
>cost of a context switch.  Last week I saw an ad for a real time version
>of Unix with a guaranteed response time (switch to kernel to high priority
>process) of 5 milliseconds.  Now, for the sizes of register files we are
>talking about, a register save is about 1/1000th to 1/100th of the total time.
>
>  Hugh LaMaster, m/s 233-9,  UUCP ames!lamaster

I worked on a RT UNIX where I *measured* context switch times of 250 microsecs.
These were typical minimums, but not guaranteed.
On this process memory accesses were at least 600 ns, so saving 16 registers
was at least 3%. Not to mention that the first register save was nearly always
a TLB miss.