How can I tell if I need more memory?

[Philip Herlihy]

I have a laptop running XP SP2, with 768Mb of main memory (I've always
thought this was a lot!). I've played around with the various performance
counters, especially the memory ones, but I'm struggling to interpret the
results. What's the surest way to assess whether more memory would speed
things up?

 

[Alan Smith]

I have a laptop running XP SP2, with 768Mb of main memory (I've always
thought this was a lot!). I've played around with the various performance
counters, especially the memory ones, but I'm struggling to interpret the
results. What's the surest way to assess whether more memory would speed
things up?

It is best judged when using the computer normally. Run your program, do
what you normally do. Does it seem slow? Playing with performance counters
won't make the computer faster.
If you have a specific program or use that you have a problem with then ask.

 

[David Candy]

The amount of swapping. Not the amount of swap file used.


An example of the second. The Add Font dialog. Noone uses it. If your computer isn't rebooted for a year it will soon end up in the swap file (though being in an dll it will be swapped to the dll it lives in - ie as dll don't change it will be discarded and read if needed - all open executable files become swap files which is why you can't delete them). Many things will end up swapped. This is not a problem. It merely getting rid of unused memory so it can give it to something that might use it.

However data that you use frequently should stay in memory most of the time. It's frequent swapping in and out that is the problem.

I doubt you have any memory problem. You are most likely to have too much memory which is merely a waste of your cash..

 

[Bert Kinney]

Hi Philip,

If you read the "How big should the page file be" section in the
following article you will find a formula for detecting if your system
will benefit from more RAM.
Virtual Memory in Windows XP
http://aumha.org/win5/a/xpvm.php

 

[Philip Herlihy]

Thanks - I'll study that carefully! There are lots of "rules of thumb" but
I'm looking for some real evidence!

 

[Philip Herlihy]

The article certainly made me look at my pagefile - it's set to start at
1.2Gb and doesn't seem to get any bigger than that (not surprisingly). The
control panel dialogue shows that as the "recommended" size.

With a 2GHz processor and 768Mb of memory I'd have expected the machine to
fly, even though I work it very hard (both apps and services). I use it for
web-design and graphics editing, with loads of things going on in the
background. If it was paging I'd expect an occasional delay corresponding
to a sudden burst of activity on the disk light, after which it would "fly"
again, but although it seems to be pretty quick for most operations I do get
occasional slowdowns, with either the processor or the disk light maxed out.
I'd like to use the performance monitor to understand it better, but I'm
having trouble interpreting the results.

 

[Anna]

The article certainly made me look at my pagefile - it's set to start at
1.2Gb and doesn't seem to get any bigger than that (not surprisingly).
The control panel dialogue shows that as the "recommended" size.

With a 2GHz processor and 768Mb of memory I'd have expected the machine to
fly, even though I work it very hard (both apps and services). I use it
for web-design and graphics editing, with loads of things going on in the
background. If it was paging I'd expect an occasional delay corresponding
to a sudden burst of activity on the disk light, after which it would
"fly" again, but although it seems to be pretty quick for most operations
I do get occasional slowdowns, with either the processor or the disk light
maxed out. I'd like to use the performance monitor to understand it
better, but I'm having trouble interpreting the results.

Philip:
On the face of things it would seem that your present 768 MB of RAM should
be more than sufficient for your purposes, but on the other hand since
you're working with some relatively high memory-intensive applications, an
extra 256 MB of memory to bring you up to 1 GB or so might possibly make
things work a bit more sprightly. It's really next to impossible to tell
until you try it. In my experience, from a cost vs. value point of view,
there are no programs/hardware/rules-of-thumb which *really* work to tell
you with any degree of precision just how much memory one needs. As a
practical matter all one can do in most cases is install additional RAM to
determine if there is any substantive improvement in the computer's
performance as a result of the increased memory.

You mention your laptop is equipped with a 2 GHz processor. While you should
be getting a reasonable amount of performance from that machine with 768 MB
of RAM, I'm not certain you can expect it to "fly", depending upon what you
mean by that. By today's standards a 2 GHz processor is rather modest and
since you're working with some high-intensity applications "with loads of
things going on in the background", it could very well be it's your
processing power that needs the upgrading. But of course, when you're
dealing with a laptop/notebook, an upgrade of the processor alone is not
normally a practical option, sad to say.
Anna

 

[Philip Herlihy]

Thanks for the comments, but it is possible to measure what's going on and
identify the bottleneck or lack of one, maybe. The key, I believe, is to
use the performance monitoring tools, but I need help interpreting the
results.

 

[Anna]

####################
## PH, London
####################

Thanks for the comments, but it is possible to measure what's going on and
identify the bottleneck or lack of one, maybe. The key, I believe, is to
use the performance monitoring tools, but I need help interpreting the
results.

Well, Phillip I certainly wish you luck finding a tool or measurement
technique that will allow the user to determine with some degree of
certainty whether adding this or that amount of RAM will result in some
performance gain that can be specifically and accurately measured. I know
I've never found such. But if & when you do, please post your findings. I
sure would like to know about it and I'm sure others would as well.
Anna

 

[David Candy]

I told him how but the ignorant pig ignores me.

 

[Alan Smith]

Well, Phillip I certainly wish you luck finding a tool or measurement
technique that will allow the user to determine with some degree of
certainty whether adding this or that amount of RAM will result in some
performance gain that can be specifically and accurately measured. I know
I've never found such. But if & when you do, please post your findings. I
sure would like to know about it and I'm sure others would as well.
Anna

And I said the similar too. It is real world performance in an application
that matters not a benchmark figure. He still hasn't said what real problem
he has, just an attempt to increase benchmark figures. Why not just try as
many benchmarks as possible then use the one that gives the biggest number?

 

[David Candy]

Given a choice between you, me, or Anna I think I'd choose Anna too.

 

[Philip Herlihy]

Thanks to (almost) everyone who has taken the trouble to contribute.

Some years ago I read a dense treatise on performance, a chapter in a
boggling book on windows internals. That discussed (in very great detail) a
baffling array of counters and their interaction. I don't have the book to
hand now, and recently found I couldn't interpret the results when I tried
running the counters (Control Panel, Administrative Tools, Performance).
So, I asked here. Bert's reply was helpful, leading me to consider (and
exclude) page-file sizing as a contribution. I value the goodwill of those
who shared their experience and outlook, but was looking for something
definite, along the lines of that half-remembered chapter. I wasn't even
sure if my expectations of performance were reasonable, of course.

Someone in another forum (uk.comp.misc) came up with a solution which is
simpler and more elegant than I'd expected. Task Manager, under the
Performance tab, gives a figure for "Commit Charge", which is described as:

"Memory allocated to programs and the operating system. Because of memory
copied to the paging file, called virtual memory, the value listed under
Peak may exceed the maximum physical memory"

It appears to be enough to compare the Peak value of Commit Charge
(currently 622,248) with Physical memory (785,648). I hadn't previously
registered what this metric denoted. These figures shows my machine hasn't
yet (today) had to swap out anything, so it seems likely that my machine's
limits are within the disk, CPU or some other less easily-upgraded
component. I might be able to save the £90 I was going to blow on another
Gb of memory. On the other hand, I'm not running Dreamweaver or Photoshop
(for example) yet, so I'll monitor the situation over the next few days.
Now, with Dreamweaver, Photoshop and FrontPage loaded, and a file open in
each, my Commit Charge is now 726,444. Looks like I have enough memory!
Perhaps it's my expectations that need adjusting...

 

[Guest]

Performance monitoring is a very tricky task, and I love that you have
attacked it with the perfmon tool.

Too many people have no idea what that tool really can tell you becayse of
the dizzying array of counters and objects.

Lets lay some groundwork first.

Going in assuming that one area is a potential bottleneck will skew your
search, so go in open minded.

I will assume you know the diffrence between objects, counters, and instances.

There are four basic items that will effect performance:
CPU (number and speed)
RAM (amount and type/speed)
Disk access speed
Network interface

To get any meaningful results you should monitor at least the first three.
As an initial baseline I would monitor the

Processor : percent of CPU usage
Server work queues: Que Length
Memory : available mbits
Memory : Pages per sec
Pagefile: % usage
Physical disk : current Que length

A combination of high CPU activity, low pages per second on Memory, Low
available Mbits, and high % usage on page file indicate a bottleneck in the
RAM.

The ques let you know if there are any other bottlenecks. A server work que
length over 5 for any amount of time is indicitive of a proc bottleneck.

High disk ques show a slow disk subsystem. Combine that with high page usage
and you will see a massiv eperformance drop when excessive paging occurs.

I don't have the link, but there used to be excel macro that you could dump
raw perfmon data into (csv file) and would pinpoint bottlenecks.

 

[Philip Herlihy]

Thanks, Manny - I've already tried these counters and come to some pretty
clear conclusions. My machine is disk-bound if anything.

Thanks!