Path: utzoo!utgpu!watmath!uunet!uvm-gen!banzai!jay
From: jay@banzai.UUCP (Jay Schuster)
Newsgroups: unix-pc.general
Subject: Re: Performance questions
Keywords: performance sticky kernel tuning
Message-ID: <1049@banzai.UUCP>
Date: 21 Dec 88 14:33:11 GMT
References: <19@dons3b1.UUCP> <6412@rphroy.UUCP>
Reply-To: jay@banzai.UUCP (Jay Schuster)
Organization: People's Computer Company, Williston, VT
Lines: 25

In article <6412@rphroy.UUCP> tkacik@rphroy.UUCP (Tom Tkacik) writes:
>I tried using the sticky bit on vi at one time, and found that there was
>no decrease in load time at all.  I came to the conclusion that the sticky bit
>did not work on the Unix-PC.  That was with 1Meg of memory.  I now have 2.5Meg,
>and recently tried it again.  This time I found that the load time for vi
>went from about 5 seconds to about 2.  I have been running gdb (the GNU
>debugger), which has an executable of about 150K.  This seemed to take forever
>to load, about 11 seconds.  Setting the sticky bit brought it down to under
>4 sec.  Currently I also have the sticky bit set on the GCC-cc1, about 400K,
>and the spreadsheet sc, 120K.  All have had load times drop by less than half.
>

The sticky bit makes the program be loaded off of the swapper
instead of from the filesystem.  Since vi is part of the installed
original set (most likely), grabbing it off of probably sequentially
allocated blocks off of the filesystem versus getting it off of
the definitely sequentially allocated blocks of the swapper isn't
so much of a win.  If your filesystem is fragmented, which it
probably is when you end up installing huge things like gcc and
gdb, then setting the sticky bit has the effect of `moving' the
fragmented file into an unfragmented space, increasing load speed.

-- 
Jay Schuster				uunet!uvm-gen!banzai!jay
The People's Computer Company		`Revolutionary Programming'