[vox-tech] Three Install Questions

[Peter Jay Salzman]

On Mon 21 Feb 05, 10:00 AM, Josh Parsons <jbparsons at ucdavis.edu> said:
> Hi, Pete,
> 
> As you say, obviously adding more RAM to a machine reduces (rather than
> increasing) the need for swap. It's odd that people seem to behave as if
> the reverse were true. :)

:)  It's a persistant paradigm that was probably started by a person who
wrote the Red Hat 5.1 Users' Guide who was an old out of work PDP-11
administrator.

> However, I think SIDR remains a good rule of thumb for calculating how
> much swap to set up in a typical desktop system initially. I've found
> that with swap = RAM, I can run out of VM in non-pathological ways
> (unless you count running memory hog office apps as pathological). Given
> how hard it is to increase the swapsize on typical PC systems, it's
> safest to err on the side of liberality. SIDR makes for just slightly
> more VM than I'll ever need. Or so it has seemed to me.
> 
> What your point shows is that it's a mistake for someone to go upgrading
> swap just because they've installed more RAM. The only reason to
> increase swap is if they find themselves running out of VM.
 
Hi Josh,

I think we're pretty much in agreement in nearly everything.  I actually
wrote a lot more in that post, but deleted a bunch of stuff because I didn't
want to dilute the importance of my main thesis.  You mentioned some of what
got cut.  :)

The only place where I disagree with you is how SIDR is a good rule of thumb
for an *initial* swap size.  The reason why I disagree is that truly modern
systems, even home systems, are coming with a GB of RAM these days.

My main workstation is an aging dual celeron 333 whose CPUs seem to have
lost the ability to overclock reliably and with 128MB.  The system is an
over achiver in every sense.  It shouldn't be performing this well; the mobo
was designed superbly.  Despite that, the system is being pushed to its
limits every day, and is memory starved.  The reason why I don't buy more
memory is that single density DIMMs are REALLY expensive and I'll be getting
a dual opteron in the next few months, so I don't want to sink any more
money into it.

The point is, back when I initially built the system, I think my choice of
SIDR was a good one.  I never needed that much swap (to my knowledge), but
heck, who really cares about 256MB of disk space?

But these days, people are getting systems with multi memory bus channels
and putting a GB of RAM on each channel.  Your typical P4 or Athlon 3200+ is
probably going to come with 512MB of RAM or more.   SIDR is now talking
about a GB of swap.

So we're in complete agreement over everything except that I believe SIDR
was a good "starting point", not based on any need other than the need of
picking SOME number to begin with.  But that was years ago.

So to recap, my belief is:

1. SIDR was necessary pre-Linux.
2. SIDR provided a nice starting point a few years ago.
3. Today, I consider it useless for bleeding edge systems.  It outlived
   its usefulness.

Some things I cut out of my previous post:

Your ears and eyes are the best determining factors for how much swap you
need.  Multi-user system administrators need to follow a different strategy,
but for home use, we all know our use-habits.  If I added an infinite number
of DIMMs to this computer, my use-habit wouldn't really change much.

If you don't ever hear the hard drives swapping, then don't add more swap.

Lastly, here's a fun thing to try:

1. Boot up.
2. Start X.
3. Start an xterm
4. Do:  "watch cat /proc/swaps".
5. Start running applications that you normally run, one by one.
6. After each time you run a new app, look and see how much swap you're
   using.


Wilson, it's silly to make partitions for future swap just because you plan
on adding RAM to the system.  Don't do it.   It's like rebooting a Linux
system to solve a problem.  Wrong paradigm for what's sitting on your desk.
If you were a real system admin, planning on adding users to a multi-user
system, that would be one thing.  But if you're a home user with only a wife
and kids on your system, it just makes no sense.

Pete

-- 
Save Star Trek Enterprise from extinction: http://www.saveenterprise.com

GPG Fingerprint: B9F1 6CF3 47C4 7CD8 D33E  70A9 A3B9 1945 67EA 951D