[kde-linux] VM and Swap problems

[Randy Kramer]

On Sunday 28 January 2007 03:48 am, James Richard Tyrer wrote:
> My system has developed an odd problem.

Wonderful!  ;-) Well, sort of--I've had a similar problem for a long time with 
no good resolution--more below.

> I have about 1 GByte of disk swap space and 7/16 GByte of RAM.

Just to clarify, I guess you mean 7/16ths of a GB? (like 512 + ~83 MB or 
such)?

> The VM and swap work fine till the swap space is half full.  But, after 
> swap is half full the system starts removing swap until it is exactly 
> half full.  When this happens, the VM system starts to thrash and the 
> system won't do anything till the thrashing stops.

I've had a problem like this through quite a few versions of Mandrake/Mandriva 
(I can't recall when I first noticed it--it may have been as early as 
Mandrake 6.2.)  I've always used kde, but I don't think it's a kde problem.  
(Can I prove that--probably not really--I do use other applications.)

The workarounds that I am aware of:

   * Add more RAM

   * As suggested by someone else, use top or similar and when either you 
notice swap approaching half full, or you notice a slow down, start stopping 
and then restarting processes that are using a lot of memory.  (I restart 
net-monitor, any browser other than konqueror, konqueror, and kmail 
regularly, in about that order.  (I use browsers other than konqueror for 
sites that don't work with konqueror.)  (All of these applications seem to 
have what I will call memory leaks--they gradually use more and more memory 
the longer you use them and don't return it, for example, when you close a 
web page or tab.)  I base my decision on VIRT usage, although lots of people 
feel that is not a very good indicator of "real" memory use.

   * Another possible workaround might be to drastically increase the amount 
of swap, so it is very difficult to reach the 50% usage figure.   (I 
originally couldn't decide whether my thrashing / slow response problems were 
at the 50% of swap figure, or swap usage ~= available RAM (because in most 
previous installations I installed (per the rule of thumb) swap equal to two 
times RAM.  In my most recent install, my swap is 3x RAM, and the slowed 
response still occurs at 50% swap usage.  (I'm not 100% sure the problem is 
thrashing, but it is as good a guess as any--the disk does see a lot of 
activity at this point.)

Some other things I do:
   * run as little as possible especially servers--for example, I do not run 
an MTA

Some other possible contributing causes (IMHO):
   * I often wonder if this problem is partially a result of the (seemingly) 
Linux philosophy that memory not in use is wasted memory?  (For other 
reasons, I'd like to have a pool of memory that is not in use at any 
particular time, for example, so that when I go to start a new app, memory is 
immediately available for it.)
   * Similarly, I wonder if this problem is partially a result of the Linux 
design of having one kernel (and memory handling philosophy, etc.) for both 
workstations and servers?  I'd really prefer to have memory handling 
customized for my use as a workstation rather than any bias that might exist 
towards handling it (memory) for a server.  

Another thing I don't like about Linux and its use of memory is (maybe it is 
mentioned above--maybe it is the unused memory is wasted memory philosophy) 
is the way it keeps expanding.  

Several times over the years I've drastically increased the amount of RAM 
memory in my systems (I mean like doubling, first, iirc, from 8 to 16 MB ;-) 

Each time I did that I initially did not change the distro (revision) I was 
using (so I could compare memory / swap behavior before and after the 
addition of RAM)--every time, with the same mix of applications, the RAM 
memory in use and the swap usage increased to fill the available memory to 
about the same percentage as before.  In some cases I did notice improved 
performance, in other cases it was so minor as to be undetectable (to me).

Darn, I was going to mention another point but it seems to have slipped my 
mind.

Oh, wait, it had something to do with letting me have tools to control (or 
influence) the use of memory in *my* system.  Oh, and oh yes, I'm aware of 
and have experimented with swappiness.  That didn't seem very helpful to 
me--somewhere I have some notes, but it seemed that I either maintained about 
the same performance as before (i.e., the slowdown at 50% of swap in use), or 
started having crashes.  (Around this time I was using systems with 512 to 
768 MB--the most my motherboards could handle.)

But the influence I'd like to have is a little different.  I'll have to think 
about it to try to articulate it.  When I do, I'll write again.

Anyway, at least now there are two of us who have seen problems along this 
line.  (Maybe if we both write to Linus he will do something to make things 
better ;-)

Randy Kramer