--------

On Tue, 15 Jul 1997 05:30:06 -0500 (EST), 
   "John S. Dyson" <toor@dyson.iquest.net> writes:


>> You haven't lived until you've TeXed large documents using a four-page
>> resident set :).

>EEEK!!!  Why do that???? :-).


As well as the performance questions (how well did this now-dead VM
system stand up under really heavy load?), and the software-managed
TLB (it was a mips cpu) we wanted to be _really_ sure there weren't
any cache-consistency bugs.  (Someone nameless had earlier ``fixed'' a
stale-icache-footprint problem by adding free pages to the tail of a
freelist rather than the head. *Ouch*.)


We could've deliberately varied the effective process TLB resident-set
size and compared timings against time with an ``infinite'' pagecache
(420Mbytes RAM), with code and TeX inputs preloaded from an earlier
run) to measure the per-page overhead in that VM system; I don't
remember.