[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde
Subject:    Re: What can be done to accelerate KDE 1.
From:       Calvin Bruce Ostrum <cbo () interlog ! com>
Date:       1999-02-18 18:13:48
[Download RAW message or body]

At 18:02 on Thursday February 18 +0100, Mario Weilguni wrote:

| >| They donīt, I measured with kpm:
| >| kpm: SIZE=2948, RSS=2948, SHARE=2308
| >| 
| >| This means: from 2948 kb the application uses, 2308 are shared (or can be
| >| shared) with other applications. So kpm is around 600KB in size, far less than
| >| the 5 mb you said.
| >
| >How do you figure that?  When code is shared there is no
| 
| Simple. Between all KDE applications at least the following libraries are
| shared:

Yes, but the shared code is still used, and someone has to be
considered responsible for it.

| libc is shared by (nearly) all programs, all other are shared between ALL KDE
| programs. So you can say: "Application X uses VmSize-VmLib bytes".

It doesn't follow logically that sharing a library implies sharing
code.  Processes could use different routines from those libraries,
although no doubt in typical cases  many of the same routines are in fact
shared.

| >One can imagine a situation where processes shared code
| >in pairs, so that as a crude estimate, the shared portion
| >of the code took up half as much as summing the contents
| >of the SHRD column would suggest.  If this were true of kde,
| >kde would still be taking up a hell of a lot of memory.
| 
| Of course you can compute the size this way, but thatīs very strange

I don't see how this would be strange, given the conditions
I specified, which do not apply to KDE.

| suppose
| you start a KDE-application A , the you can say it takes 3 MB. And now you start
| another application B, now A takes ~1900 KB and B take 1900 KB. Now start C,
| so A, B and C each take ~1530 KB and so on.

No, in the situation I described, this would not happen.  A and B
would always be taking the 1900 no matter how many D,E,F etc were
run, since "processes shared code in paris" in that situation.  
As I pointed out, the KDE situation is *not* like this, but much 
more like what you describe.

| So I consider all these libraries as part of the operating system (another
| sort of "shared" library), and do not count them. This may not be
| completly correct, but I think itīs the best way to measure sizes.

I don't think so.  If a person has no choice but to use KDE, then
it might make sense, since this shared memory could not be recovered.
However, if she is considering whether to use KDE or not, the total
amount of memory used by a typical KDE setup (and hence, the
average amount used by a typical single KDE process) is
quite relevant.  In your scenario, you would ignore this entirely,
and that is not a fair way to compare various options open to a
person using Linux.  

Also, I don't think it is really very correct to assume that
once a piece of code is shared by only two processes, that it
is effectively being shared by all processes, so that it might
as well be considered part of the kernel OS.  It would be nice
if one could determine the structure of code sharing more 
precisely, to see where it lies between the degenerate situation
I described above, and the situation where there is far more
extensive sharing. 

---------------------------------------------------------------------------
Calvin Ostrum                                              cbo@interlog.com
---------------------------------------------------------------------------
People make their own history, but they do not make it just as they please;
they do not make it under circumstances chosen by themselves, but under
circumstances directly encountered, given and transmitted from the past.
        -- Karl Marx, "The Eighteenth Brumaire of Louis Bonaparte"
---------------------------------------------------------------------------
--
Send posts to:  kde@lists.netcentral.net
 Send all commands to:  kde-request@lists.netcentral.net
  Put your command in the SUBJECT of the message:
   "subscribe", "unsubscribe", "set digest on", or "set digest off"
PLEASE READ THE ARCHIVED MESSAGES AT http://lists.kde.org/ BEFORE POSTING
**********************************************************************
This list is from your pals at NetCentral <http://www.netcentral.net/>

[prev in list] [next in list] [prev in thread] [next in thread]