Memory Resource Management

From: Alvin (replay_at_to_the_newsgroup.thanks)
Date: 09/01/05 

Date: Thu, 01 Sep 2005 12:25:56 GMT

Hello all,

I have a program that links to a static library. I am trying to determine
how long a particular routine takes to execute from the static library. To
do this I am using gettimeofday() to calculate the elapsed time.

What I have noticed (and somewhat expect) is that each consecutive time I
run the program, the elapsed time is shorter and shorter till a point that
it just remains relatively constant. I expect that there is some caching
occurring on the OS side (SuSE 9.3 Pro - kernel 2.6.11.4-21.8-default).

However, what I find strange is that this "caching" crosses applications?
For example, I have a second program that links to the same static library.
What I have done is hacked the program to calculate the elapsed time in the
exact same way I do in the first program (used a couple of gettimeofday
calls, etc.)

What I have noticed is that if I run the first program, I get that initial
lag. However, if I then run the second program, the elapsed time is
minimal. I have also done this test in opposite order (after I run
OpenOffice and FireFox to sort of reset the scheduler and memory resource
manager) and have experienced the same thing. The 2nd program will have the
lag while the 1st will then run smooth.

What I think is happening is that in the routine I am timing memory is
allocating various arrays and data structures. The OS's memory manager then
must find the appropriate regions to accommodate the requested size, etc.
That would account for the initial lag. Then on the second run, the
allocations occur again but the OS's memory manager just reuses the same
spots in memory. Is this basically correct? If so, can this occur between
two separate processes? Or is this happening only because I am timing the
same routine from the same static library?

Any explanations, hints, tips or URLs where I can read up more on this would
be appreciated.

Thanks,

Alvin

Relevant Pages

  • Re: IOCP touching my OVERLAPPED after my Write has completed
    ... Your code will leak memory in this case. ... WSASendTo() it seems a bit different. ... uncommitted by the completion routine. ... FreeRequest(req); //in debug I decommit the memory of this request ...
    (microsoft.public.win32.programmer.kernel)
  • Re: Handling SSC interrupts in Applesoft BASIC
    ... The Apple 2 has what is called "memory mapped I/O". ... One way is to POKE values into selected ZERO PAGE locations. ... Where to put the routine? ... You could assemble it by hand! ...
    (comp.sys.apple2.programmer)
  • Re: [PATCH] x86_64/__iowrite32_copy: dont use string move for PIO writes
    ... is little point in writing this in assembly at all. ... does the I/O operations 32 bits at a time, ... The point of the routine is to make sure that transfers are *at least* ... which you better not if you're accessing WC memory. ...
    (Linux-Kernel)
  • Re: Efficient memory truncation of a matrix
    ... The whole point of this mex routine is to modify an array in place since you haven't got the memory to make a copy. ... mwSize j, M, N, Mnew, sadd, tadd; ... int i, ElementSize; ...
    (comp.soft-sys.matlab)
  • Re: Pointer To A Vector Elements While Still Adding
    ... > Vector elapsed time is 2724 ... Have you tried reversing the order: vector first and then deque? ... Your computer has more memory than mine, ... for a vector of final capacity pow. ...
    (comp.lang.cpp)