[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

fseek is inadequately slow (4 times slower then fread)

From: Khamenia, Valery
Subject: fseek is inadequately slow (4 times slower then fread)
Date: Tue, 23 Sep 2003 11:00:51 +0200

Hi all and especially drepper, aj and rolland.

  the random access in non-cacheable files is very slow.
  I've used the test below for benchmarking this issue 
  and find an interesting fact.

  My experiment was like following. 

  First run this test like that:

  % time -v thistest non_cached_big_file1 16 4000 16

  (e.g. the wall clock time I got is 1 min 15 sec for reading 
  1Gb file)

  Let's comment out fseek now and run with the same size 
  noncached file (or just clear the cache if you are 
  confident with the Linux internals and use the same file).

  % time -v thistest non_cached_big_file2 16 4000 16

  (and I got 30 sec for reading 1Gb file. BTW, that is about 
  33Mb/sec even with 16 bytes reading)

  Well, what do we have then? The ratio is easy to calculate,
  the system of equations is like that: 
  x+y=75sec & 2*x = 30sec => y:x = 1:4
   'x' is relative time for fread and 
   'y' is a relative time for fseek

  Thus, we see that fseek is about 4 times slower then fread.

      I got glibc 2.2 only to test.
      However the similar results were obtained with last cygwin
      distribution. Hm, it's difficult to say what glibc is used 
      there, but drepper should know I guess :-)

thanks a priori for any answers!

// -------------- TEST BEGIN
#include <cassert>
#include <cstdio>
#include <cstdlib>
#include <iostream>

using namespace std;

main(int argc, char**argv) {
  clog << "usage: "<<argv[0]<<" filename readby vbufsize fseekgap\n";
  long sz = atol(argv[2]);
  long bufsz=atol(argv[3]);
  long fseekgap = atol(argv[4]);
  char *data = new char[sz];
  char *buf = new char[bufsz+8]; // as show docs +8 is needed :-/
  FILE* f=fopen(argv[1], "rb");
  assert(!setvbuf(f, buf, _IOFBF, bufsz));
  clog << "reading by " << sz << " bytes; buffer size = "<<
     bufsz << "bytes... ";
  long i=0;
  while( fread(data, sz, 1, f) ) {    
    fseek(f, fseekgap, SEEK_CUR);
  clog << "done\n\n";
  clog << "fread is called: " << i << endl;
// -------------- TEST END  

kind regards,
Valery A.Khamenya
Bioinformatics Department
BioVisioN AG, Hannover

reply via email to

[Prev in Thread] Current Thread [Next in Thread]