Aucbvax.5624
fa.unix-wizards
utzoo!decvax!ucbvax!unix-wizards
Sun Dec 27 17:44:37 1981
V6 vs V7 swapping
>From decvax!cwruecmp!ordy@Berkeley Sun Dec 27 17:30:57 1981
A few weeks ago I sent an item similar to this one. It
apparently got lost in some mail/news/uucp problems, so here it goes
again.
On vanilla V6, it is required that the text and data segments
of a process come back into core (swap) physically contiguous.
On vanilla V7, when both need to come back, they can come
back into two disjoint pieces of memory, which should let alot more
processes be core resident, and reduce core shuffling.
My question of a few weeks ago asked if anybody had any comments
on what effect this change made on system throughput.
Since then I have added the V7 feature to one of our V6 systems.
I also had the largest piece (text or data) come in first, which the
comment in the V7 code suggests should be done, but was not.
The statistics I added indicate that of the complete swapins
(text + data) between 1/4 and 1/3 can be fit (and on our system now are)
only if broken up (I attempt to put them together if I can).
Attempts to load the system and get a seat of the pants feel
show that the system seems a bit more responsive.
Any comments on this area would be appreciated. We have a solid-state
swapping 'disk', and are also interested in the validity of the 3 and
2 second outage/inage values.
Greg Ordy
-----------------------------------------------------------------
gopher://quux.org/ conversion by John Goerzen <
[email protected]>
of
http://communication.ucsd.edu/A-News/
This Usenet Oldnews Archive
article may be copied and distributed freely, provided:
1. There is no money collected for the text(s) of the articles.
2. The following notice remains appended to each copy:
The Usenet Oldnews Archive: Compilation Copyright (C) 1981, 1996
Bruce Jones, Henry Spencer, David Wiseman.
