Subject: Re: tmp in mfs and swap
To: Peter Galbavy <peter@wonderland.org>
From: Phil Knaack <flipk@ncremp.ag.iastate.edu>
List: current-users
Date: 02/06/1996 11:33:50
>Using -pipe is a big win. On a system with not enough RAM, and not using
>-pip, the stuff gets written to intermidiate files. If this is an MFS /tmp
>then this is swap, else it is ordinary file system.

Actually no, if its MFS, its the user space of the mfs process. If there's
not enough space, then its in swap.

>On a system with less
>memory and -pipe the processes get swapped. If the swapper/pager is good
>enough then what is the difference ?

Probably not much; both involve a lot of context switching (with -pipe
you're switching between cc1 and as, with MFS tmp you're switching between
cc1 and mfs).

>On a system with lots of mem -pipe is a major winner. My default editing
>procedure for config generated Makefiles.

Actually you might find this interesting. 

Last night I brought blade.exnet.iastate.edu (486slc2/66 with 16M ram) down
to single-user mode, mounted / and /usr, and started building kernels,
preceeding each one with a "cat /*bin/* > /dev/null" to flush caches and
whatnot, and competely blowing away the compile/BLADE directory each time.

I did one kernel with -pipe, one with ffs tmp, and one with mfs tmp.

All three took on the order of ~56 minutes. (I have a lot of stuff in my
kernel, and the Cyrix SLC processor is a 16-bit-wide kludge just like the
386sx.)

I don't have the numbers in front of me (for some odd reason I wrote them 
down on a sheet of paper rather than into a file [!]), but placing the three
runs in order of fastest to slowest:

	1. MFS tmp
	2. -pipe
	3. FFS tmp

I can post later with the actual numbers, but in a ~56 minute compile, the
difference between the first and last was ~40 seconds or so.

Perhaps the gaps will be much wider or in a much different order on a 
machine without 16-bit-multiplexing of a 32-bit bus?

I wouldn't be surprised if it turned out that #2 was slower than #1 only
because the SLC processor takes a big hit flushing its 16k cache during 
context switching..

Cheers,
Phil
--
Phillip F Knaack               flipk@iastate.edu
Database Programmer, NCREMP    Student Development Group
ISU Extension                  Project Vincent, Iowa State University