Degraded performance in cat + patch

[Tzvi Rotshtein]

Hi,
I've been using "cat" to feed large files into some data cruncher
application using something like this:
   cat my_data | data_cruncher

However, cat was reading/writing the file in sub-optimal speeds (not even
half as fast as the disk & os can provide it). I traced this to the buffer
size selection algorithm in "cat", while generally provides good balance
with low memory footprint, it constraints cat from reaching the disk's (or
OS caches) peak performance.

While it is usually not crucial for most applications to have "cat"
operating at peak performance, I thought it would be useful to let the user
determine that.
I have made the following changes to allow the user provide and override the
buffer sizes in "cat", effectively improving performance. Here's a quick
benchmark (it's a typical result after multiple runs on Linux x86 with
kernel 2.6.18):


$ time ./cat test_sample_150mb_file.txt > /dev/null
0.00user 0.54system 0:00.59elapsed 90%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+186minor)pagefaults 0swaps

$ time ./cat -r 1048576 test_sample_150mb_file.txt > /dev/null
0.00user 0.09system 0:00.12elapsed 73%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+444minor)pagefaults 0swaps


The ability to specify an explicit (and larger) buffer size has improved the
performance by a factor of x5 on my test system, which is quite a noticeable
gain, especially when dealing with files at least 50GB in size.

Let me know what do you think of it. The patch I used is available below.

-- Tzvi



*** coreutils-7.1/src/cat.c 2008-12-21 08:13:31.000000000 -0600
--- coreutils-7.1-new/src/cat.c 2009-03-05 22:52:47.000000000 -0600
***************
*** 40,45 ****
--- 40,46 ----
  #include "quote.h"
  #include "safe-read.h"
  #include "xfreopen.h"
+ #include "xstrtol.h"

  /* The official name of this program (e.g., no `g' prefix).  */
  #define PROGRAM_NAME "cat"
***************
*** 102,107 ****
--- 103,109 ----
    -e                       equivalent to -vE\n\
    -E, --show-ends          display $ at end of each line\n\
    -n, --number             number all output lines\n\
+   -r, --read-buffer-size   enforce a read buffer size (in bytes)\n\
    -s, --squeeze-blank      suppress repeated empty output lines\n\
  "), stdout);
        fputs (_("\
***************
*** 109,114 ****
--- 111,117 ----
    -T, --show-tabs          display TAB characters as ^I\n\
    -u                       (ignored)\n\
    -v, --show-nonprinting   use ^ and M- notation, except for LFD and
TAB\n\
+   -w, --write-buffer-size  enforce a write buffer size (in bytes)\n\
  "), stdout);
        fputs (HELP_OPTION_DESCRIPTION, stdout);
        fputs (VERSION_OPTION_DESCRIPTION, stdout);
***************
*** 546,551 ****
--- 549,556 ----
    bool show_ends = false;
    bool show_nonprinting = false;
    bool show_tabs = false;
+   size_t read_buffer_size = 0;
+   size_t write_buffer_size = 0;
    int file_open_mode = O_RDONLY;

    static struct option const long_options[] =
***************
*** 557,562 ****
--- 562,569 ----
      {"show-ends", no_argument, NULL, 'E'},
      {"show-tabs", no_argument, NULL, 'T'},
      {"show-all", no_argument, NULL, 'A'},
+     {"read-buffer-size", required_argument, NULL, 'r'},
+     {"write-buffer-size", required_argument, NULL, 'w'},
      {GETOPT_HELP_OPTION_DECL},
      {GETOPT_VERSION_OPTION_DECL},
      {NULL, 0, NULL, 0}
***************
*** 576,582 ****

    /* Parse command line options.  */

!   while ((c = getopt_long (argc, argv, "benstuvAET", long_options, NULL))
   != -1)
      {
        switch (c)
--- 583,589 ----

    /* Parse command line options.  */

!   while ((c = getopt_long (argc, argv, "benr:stuvw:AET", long_options,
NULL))
   != -1)
      {
        switch (c)
***************
*** 595,600 ****
--- 602,616 ----
    number = true;
    break;

+ case 'r':
+   {
+               long int buffer_size;
+       if ( xstrtol (optarg, NULL, 10, &buffer_size, "") != LONGINT_OK )
+  error (EXIT_FAILURE, 0, _("%s: invalid read buffer size"), optarg);
+               read_buffer_size = buffer_size;
+   }
+   break;
+
   case 's':
    squeeze_blank = true;
    break;
***************
*** 612,617 ****
--- 628,642 ----
    show_nonprinting = true;
    break;

+ case 'w':
+   {
+               long int buffer_size;
+       if ( xstrtol (optarg, NULL, 10, &buffer_size, "") != LONGINT_OK )
+  error (EXIT_FAILURE, 0, _("%s: invalid write buffer size"), optarg);
+       write_buffer_size = buffer_size;
+   }
+   break;
+
   case 'A':
    show_nonprinting = true;
    show_ends = true;
***************
*** 641,646 ****
--- 666,674 ----
      error (EXIT_FAILURE, errno, _("standard output"));

    outsize = ST_BLKSIZE (stat_buf);
+   if ( write_buffer_size > 0 )
+      outsize = write_buffer_size;
+
    /* Input file can be output file for non-regular files.
       fstat on pipes returns S_IFSOCK on some systems, S_IFIFO
       on others, so the checking should not be done for those types,
***************
*** 705,710 ****
--- 733,740 ----
    goto contin;
   }
        insize = ST_BLKSIZE (stat_buf);
+       if ( read_buffer_size > 0 )
+           insize = read_buffer_size;

        /* Compare the device and i-node numbers of this input file with
   the corresponding values of the (output file associated with)