views:

275

answers:

5

I am trying to understand why grep built by me is much slower than the one that comes with the system and trying to find what compiler options are used by grep that comes with the system.

OS Version: CentOS release 5.3 (Final) grep on system:

  Version: grep (GNU grep) 2.5.1
  Size: 88896 bytes
  ldd output: 
 libpcre.so.0 => /lib64/libpcre.so.0 (0x0000003991800000)
 libc.so.6 => /lib64/libc.so.6 (0x0000003985a00000)
 /lib64/ld-linux-x86-64.so.2 (0x0000003984a00000)

grep built by me:

  Version: 2.5.1
  Size: 256437 bytes
  ldd output:
 libpcre.so.0 => /lib64/libpcre.so.0 (0x0000003991800000)
 libc.so.6 => /lib64/libc.so.6 (0x0000003985a00000)
 /lib64/ld-linux-x86-64.so.2 (0x0000003984a00000)

The performance of system grep (330 msecs) is way faster than grep that I built (22430 msecs) when run a regex search on a large list text file.

Following is the command I used to time ..

% time src/grep ".*asa.*" large_list.txt > /dev/null
real 0m22.430s
user 0m22.291s
sys 0m0.080s

OR

% time bin/grep ".*asa.*" large_list.txt > /dev/null
real 0m0.331s
user 0m0.236s
sys 0m0.081s

The system grep is clearly using some optiomizing options that is giving huge performance difference.

Can some body help me with what options the system grep may be built with?

Here is the compile options for one of the source files when I build ..
gcc -DLIBDIR=\"/usr/local/lib\" -DHAVE_CONFIG_H -I. -I.. -I.. -I. -I../intl -g -O2 -MT xstrtol.o -MD -MP -MF .deps/xstrtol.Tpo -c -o xstrtol.o xstrtol.c

The output of ./configure:

checking for a BSD-compatible install... /usr/bin/install -c
checking whether build environment is sane... yes
checking for a thread-safe mkdir -p... /bin/mkdir -p
checking for gawk... gawk
checking whether make sets $(MAKE)... yes
checking build system type... x86_64-unknown-linux-gnu
checking host system type... x86_64-unknown-linux-gnu
checking for gawk... (cached) gawk
checking for gcc... gcc
checking for C compiler default output file name... a.out
checking whether the C compiler works... yes
checking whether we are cross compiling... no
checking for suffix of executables... 
checking for suffix of object files... o
checking whether we are using the GNU C compiler... yes
checking whether gcc accepts -g... yes
checking for gcc option to accept ISO C89... none needed
checking for style of include used by make... GNU
checking dependency style of gcc... gcc3
checking for a BSD-compatible install... /usr/bin/install -c
checking for ranlib... ranlib
checking for getconf... getconf
checking for CFLAGS value to request large file support... 
checking for LDFLAGS value to request large file support... 
checking for LIBS value to request large file support... 
checking for _FILE_OFFSET_BITS... no
checking for _LARGEFILE_SOURCE... no
checking for _LARGE_FILES... no
checking for function prototypes... yes
checking how to run the C preprocessor... gcc -E
checking for grep that handles long lines and -e... /bin/grep
checking for egrep... /bin/grep -E
checking for ANSI C header files... yes
checking for sys/types.h... yes
checking for sys/stat.h... yes
checking for stdlib.h... yes
checking for string.h... yes
checking for memory.h... yes
checking for strings.h... yes
checking for inttypes.h... yes
checking for stdint.h... yes
checking for unistd.h... yes
checking for string.h... (cached) yes
checking for size_t... yes
checking for ssize_t... yes
checking for an ANSI C-conforming const... yes
checking for inttypes.h... yes
checking for unsigned long long... yes
checking for ANSI C header files... (cached) yes
checking for string.h... (cached) yes
checking for stdlib.h... (cached) yes
checking sys/param.h usability... yes
checking sys/param.h presence... yes
checking for sys/param.h... yes
checking for memory.h... (cached) yes
checking for unistd.h... (cached) yes
checking libintl.h usability... yes
checking libintl.h presence... yes
checking for libintl.h... yes
checking wctype.h usability... yes
checking wctype.h presence... yes
checking for wctype.h... yes
checking wchar.h usability... yes
checking wchar.h presence... yes
checking for wchar.h... yes
checking for dirent.h that defines DIR... yes
checking for library containing opendir... none required
checking whether stat file-mode macros are broken... no
checking for working alloca.h... yes
checking for alloca... yes
checking whether closedir returns void... no
checking for stdlib.h... (cached) yes
checking for unistd.h... (cached) yes
checking for getpagesize... yes
checking for working mmap... yes
checking for btowc... yes
checking for isascii... yes
checking for iswctype... yes
checking for mbrlen... yes
checking for memmove... yes
checking for setmode... no
checking for strerror... yes
checking for wcrtomb... yes
checking for wcscoll... yes
checking for wctype... yes
checking whether mbrtowc and mbstate_t are properly declared... yes
checking for stdlib.h... (cached) yes
checking for mbstate_t... yes
checking for memchr... yes
checking for stpcpy... yes
checking for strtoul... yes
checking for atexit... yes
checking for fnmatch... yes
checking for stdlib.h... (cached) yes
checking whether  defines strtoumax as a macro... no
checking for strtoumax... yes
checking whether strtoul is declared... yes
checking whether strtoull is declared... yes
checking for strerror in -lcposix... no
checking for inline... inline
checking for off_t... yes
checking whether we are using the GNU C Library 2.1 or newer... yes
checking argz.h usability... yes
checking argz.h presence... yes
checking for argz.h... yes
checking limits.h usability... yes
checking limits.h presence... yes
checking for limits.h... yes
checking locale.h usability... yes
checking locale.h presence... yes
checking for locale.h... yes
checking nl_types.h usability... yes
checking nl_types.h presence... yes
checking for nl_types.h... yes
checking malloc.h usability... yes
checking malloc.h presence... yes
checking for malloc.h... yes
checking stddef.h usability... yes
checking stddef.h presence... yes
checking for stddef.h... yes
checking for stdlib.h... (cached) yes
checking for string.h... (cached) yes
checking for unistd.h... (cached) yes
checking for sys/param.h... (cached) yes
checking for feof_unlocked... yes
checking for fgets_unlocked... yes
checking for getcwd... yes
checking for getegid... yes
checking for geteuid... yes
checking for getgid... yes
checking for getuid... yes
checking for mempcpy... yes
checking for munmap... yes
checking for putenv... yes
checking for setenv... yes
checking for setlocale... yes
checking for stpcpy... (cached) yes
checking for strchr... yes
checking for strcasecmp... yes
checking for strdup... yes
checking for strtoul... (cached) yes
checking for tsearch... yes
checking for __argz_count... yes
checking for __argz_stringify... yes
checking for __argz_next... yes
checking for iconv... yes
checking for iconv declaration... 
         extern size_t iconv (iconv_t cd, char * *inbuf, size_t *inbytesleft, char * *outbuf, size_t *outbytesleft);
checking for nl_langinfo and CODESET... yes
checking for LC_MESSAGES... yes
checking whether NLS is requested... yes
checking whether included gettext is requested... no
checking for libintl.h... (cached) yes
checking for GNU gettext in libc... yes
checking for dcgettext... yes
checking for msgfmt... /usr/bin/msgfmt
checking for gmsgfmt... /usr/bin/msgfmt
checking for xgettext... /usr/bin/xgettext
checking for bison... bison
checking version of bison... 2.3, ok
checking for catalogs to be installed...  af be bg ca cs da de el eo es et eu fi fr ga gl he hr hu id it ja ko ky lt nb nl pl pt pt_BR ro ru rw sk sl sr sv tr uk vi zh_TW
checking for dos file convention... no
checking host system type... (cached) x86_64-unknown-linux-gnu
checking host system type... (cached) x86_64-unknown-linux-gnu
checking for DJGPP environment... no
checking for environ variable separator... :
checking for working re_compile_pattern... yes
checking for getopt_long... yes
configure: WARNING: Included lib/regex.c not used
checking whether strerror_r is declared... yes
checking for strerror_r... yes
checking whether strerror_r returns char *... no
checking for strerror... (cached) yes
checking for strerror_r... (cached) yes
checking for vprintf... yes
checking for doprnt... no
checking for ANSI C header files... (cached) yes
checking for working malloc... yes
checking for working realloc... yes
checking for pcre_exec in -lpcre... yes
configure: creating ./config.status
config.status: creating Makefile
config.status: creating lib/Makefile
config.status: creating lib/posix/Makefile
config.status: creating src/Makefile
config.status: creating tests/Makefile
config.status: creating po/Makefile.in
config.status: creating intl/Makefile
config.status: WARNING:  intl/Makefile.in seems to ignore the --datarootdir setting
config.status: creating doc/Makefile
config.status: creating m4/Makefile
config.status: creating vms/Makefile
config.status: creating bootstrap/Makefile
config.status: creating config.h
config.status: config.h is unchanged
config.status: executing depfiles commands
config.status: executing default-1 commands
config.status: creating po/POTFILES
config.status: creating po/Makefile
config.status: executing stamp-h commands

Thanks, kumar

+4  A: 

You compiled with the -O2 flag. Why didn't you use the -03 flag. See here for an explanation of the optimization options available with gcc.

Using Intel's ICC compiler can also help to improve performance, though this really depends on the app. Also, it's not free.

Edit, I just saw the -g flag on your compile line. Remove that as it's turning on debug stuff and this can cause a pretty serious performance hit

Glen
Have almost the same timing with O3 ..real 0m22.271suser 0m22.163ssys 0m0.079sThe size of grep with -O3 is 344944 bytes
kumar
A: 

With that big of a performance gap, it's probably an algorithm/code difference, not just a difference in compiler optimization level. What makes you suspect the compiler?

Asher Dunn
I got the grep code from GNU (2.5.4) and shouldn't be any big change in code when compared to 2.5.1 that the system has. I am not using any of my code for running this test.
kumar
Oh, got it. I misunderstood.
Asher Dunn
+1  A: 

Another think to note besides the -O options is it looks like you are building with debugging symbols "-g".

Debug usually increases binary size and can reduce performance of said binary, I would image grep is pretty stable and you don't really need debug symbols for it.

Brian Gianforcaro
Removing -g definitely reduced the size to 101733 (with -O2), but the timing is still the same ..real 0m22.358suser 0m22.231ssys 0m0.093s
kumar
+7  A: 

Why don't you just get CentOS's SRPM for the grep binary and compare their compile options to yours? I would guess that this is much more efficient than having the entire StackOverflow community blindly poke around in the dark until they hit something.

EDIT: Are you using a locale with a multibyte encoding? (Note: if you have no idea what that means, then the answer is probably "Yes", since UTF-8 has been the default for most Linux distributions for several years now and indeed RedHat (and thus CentOS) were the very first to make the switch).

In that case, GNU grep is dog slow. And this not only applies to GNU grep but to pretty much all GNU tools that do some kind of text processing. The FSF refuses to accept any patches to improve multibyte performance, unless those patches are proven to not slow down fixed-width encodings. However, since any patch to improve performance for multibyte encodings must at least contain some if statement somewhere, it is actually impossible to write a patch that does not at least slow down fixed-width encodings by at least the overhead of that if statement. Thus, UTF-8 performance of GNU tools will continue to suck until the end of time.

Anyway, most Linux distributors don't give a rat's bleep what the FSF thinks and patch GNU grep anyway. The Fedora Rawhide SRPM contains a patch called grep-2.5.3-egf-speedup.patch, which speeds up the UTF-8 performance of GNU grep by several orders of magnitude. (Since this patch is already from 2005, I assume that it is also used in CentOS.) This patch is also used in Mac OSX, Debian, Ubuntu, ..., pretty much nobody uses GNU grep as distributed by GNU. Text processing in a multibyte encoding will never be as fast as in a fixed-width encoding, but it should at least be comparable, not 50x (or even 1500x as some people have reported) slower.

There's also another patch called dfa-optional, which makes grep simply use GNU libc's regex engine instead of its own, which is not only much faster when dealing with UTF-8 but also has far fewer bugs.

So, you might want to re-run your benchmarks with export LC_ALL=POSIX set. If that fixes your problem, you need to apply either one of the two above-mentioned patches.

More information is also available in these two RedHat bugreports:

The moral of the story: despite popular belief, the Linux distributors do know what they are doing, at least sometimes. Don't second-guess them.

Jörg W Mittag
Thanks for the excellent suggestion. I tried building from the srpm, but the performance is still slow when compared to the grep binary on the system or in the rpm. Maybe it needs to be built in a specific environment to get the right configurations? I have posted the question on one of te centos forums. I see the same behavior on other linux flavors too.
kumar
+1  A: 

What version of GCC are you using? IIRC, GCC 4 was significantly redesigned, which invalidated some of the optimization code for a while.

Max Lybbert