Thanks for the info. I figured it was more complicated than the simple summation of the memory footprint of each program. Regards, Michael Martinez ISTM/CSREES United States Department of Agriculture --- This email is signed with my digital signature so that you may verify the authenticity of the sender. --> -----Original Message----- --> From: Jan Hudec [mailto:bulb@ucw.cz] --> Sent: Sunday, January 25, 2004 12:31 PM --> To: shyam jith --> Cc: Martinez, Michael; kernelnewbies@nl.linux.org --> Subject: Re: How to calculate memory used by several processes --> --> --> On Sat, Jan 24, 2004 at 14:11:14 +0000, shyam jith wrote: --> > >>>Let's say you've got a process like inetd that forks --> off several copies --> > >>>of the same program, each of which forks several --> copies of some other --> > >>program, as follows:>>>> --> > Inted -> Program X -> Program Y --> > -> Program Y --> > -> Program X -> Program Y --> > -> Program Y --> > -> Program X -> Program Y --> > -> Program X -> Program Y --> > /* --> > So at any one time the system process table will show --> four instances of --> > Program X and eight instances of Program Y. Let's say this is on a --> > Redhat AS 2.1 system running kernel 2.4.9-e.3. --> > --> > What's the total system RAM consumed by these programs? --> Is it simply the --> > sum of the amount of RAM required by Program X and --> Program Y multiplied --> > by four and eight respectively? Or is there something --> more complicated --> > in the way the Linux kernel manages memory? --> > --> > The practical reason for the question is to determine the --> maximum number --> > of Program X that we can allow (based on the amount of --> free system RAM) --> > in order to prevent the system from swapping to disk when too many --> > Program Xs are running. --> --> Generaly, there are too many factors affecting memory --> performance. The --> best way to determine the number is probably to put a test --> load on the --> system and slowly increase the number of processes allowed until CPU --> time spent in user mode starts to drop and there cut it off. --> --> To get some qualified guess: --> - Binary files and shared libraries are loaded only once --> no matter how --> many processes use them. Moreover, they are never copied --> to swap -- --> instead, they are discarded and reloaded from the binary --> when needed. --> - The same is true for files mapped by processes using the --> mmap call. --> - Process data are private to each process. But after --> fork, they are --> not copied until they are modified. AFAIK there is no --> way to count --> pages shared in this way. --> - Files accessed by read/write will be cached. --> - Linux does not make any differnce between pages with --> different origin --> and different use. If it's accessed often, it will be in --> memory, no --> matter if it's a code, data or a cached file. --> --> ------------------------------------------------------------ --> ------------------- --> Jan 'Bulb' --> Hudec <bulb@ucw.cz> --> -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
