Re: how much memory is adequate?

[Chris Watt <cnww chebucto ns ca>]

At 07:10 02/09/00 -0400, DAN WHEELER wrote: 
>
>I have RH 6.0 installed on one of my computers and my question is how much 
>memory is required to make things run exceptionally well? During the
install I 
>selected everything as far as packages go.
>The reason I am asking is that when running X some apps take an
exceptionally 
>long time to load.
>The machine has a p233 mmx processor and 32 MB of PC 100 ram.

It depends on what you're doing. 32mb is a good, albeit minimal, start, but
unless you're running an unusually stripped-down installation you're
probably using a fair bit of swap space (AKA virtual memory) most of the time.

Basically there are three approaches you can take, depending on how much
money you have to spend on RAM:

1. Live with little memory (cheap if you do it right):
 If you've got reasonably fast hard disks (most SCSI-2 drives would
qualify, or IDE drives that are operating in DMA mode and have been
well-tweaked with hdparm) and you have split your swap space up into
equal-priority chunks on multiple separate drives (on separate controllers)
then you should be able to access swap space quickly enough (to the tune of
20 to 40 mb/sec depending on how many and what type of controllers) that it
won't be much of a problem, especially if you're not swapping to the same
drive as you're loading your applications from (although this is a tall
order on most systems).

2. Living with more, but not "enough" memory:
 As long as you're using swap space to store information swapped out of
memory for "active" programs you are not getting optimal performance out of
your system, but it doesn't take a genius to realize that the less swap
space you have to use the better. Thus if you have a system that really
"wants" 1gb of RAM, going from 32mb to 64mb will still give you a major
improvement. In my experience the improvement you get by doubling the
amount of ram in your system is fairly nearly constant (and significant)
until you hit the limit of the amount of physical memory your programs are
actually using. Then you get no further improvement. So for those with a
limited hardware-budget the saying "more ram is good" could be a mantra. If
you use software that makes "good" use of swap space, which is to say that
when it lets something get swapped out it then doesn't access it for quite
a while, you may find that you can get most of the performance gains with
much less RAM than your applications actually demand.

3. If you want "really good" performance then the only answer is to stick
in enough ram that you really do not need to use swap space. How much is
"enough"? This depends on your system, thankfully the nice folks who write
GNU software make this easy to figure out (thankyou Brian Edmonds). Start
up all the stuff you plan to be running at the same time and then go to a
console window and run the "free" command. This add up the amount of
physical memory in use (first value in the second column) and the amount of
swap space in use (last value in the second column) then divide by 1024 to
convert to megabytes. Round this value up to the nearest quantity you can
easily put in your machine (e.g. If it comes out to 117mb you'll probably
round up to 128) in terms of whatever memory your box takes (if you take
72pin SIMMs then figure in working in units of 8, 16, and 32. If you take
DIMMs then figure on units of 32, 64, 128 and 256 (512mb DIMMs are nearly
always overpriced and many boards won't support them anyway). If you find
yourself looking at amounts over 1gb, make sure you read your mainboard's
manual first. A lot of boards have a memory limit somewhere in the range
from 768mb to 2gb (AFAICT the most common is 1.5gb) and simply won't let
you use any more (in which cae see option 2 above).

p.s. In terms of X apps taking a long time to load, there is another common
problem that people migrating from Windows tend to hit with loading X
applications. When Windows is working to load an app it produces a "please
wait" cursor which, in a well-planned use of applied psychology, lets the
user know that the computer is busy and (on multiple exposures) conditions
them to wait for it to un-busy itself. Additionally large slow applications
will typically waste their first few seconds of load time by loading and
displaying a "splash screen" banner, which again gives the user the
impression that something is happening. In contrast when you start a
typical large X application your system simply loads the application as
quickly as it can, without any overt indication that it is doing so until
it is done. This has two problems, first that a new user may believe that
"clicking the button didn't work" and as such hit it again (I've seen
people open up three or four instances of Netscape while one was already
loading), and the second is that as Windows users they've been conditioned
to believe that the little pointy-arrow mouse cursor means that the system
should be ready to go do something and therefor they aren't mentally
prepared to wait. Being a psychology student I actually did some
measurements, and overall it seemed that large programs (I tested
Mathematica, Wordperfect, and Netscape) actually loaded faster in X than
they did in Windows 95 (Win95 loaded them faster than 98 or NT did, so I
used it as a benchmark) on the same system (A Celeron 500 with 256mb of ram
and a single Ultra2 SCSI hdd). This is of course not a definitive study,
since I only looked at 1 system and 3 programs.

I hope somebody actually enjoyed this ramble :) Anyone still reading at
this point, give yourself a pat on the back and then get back to real life.
--

Who is this General Failure, and why is he reading my hard disk?