grep speed on Opteron vs P4

[Terry Horsnell]

I'm seeing huge discrepancies in grep speed between
my Opteron box and P4 boxes.

The Opteron box:
----------------

Dual Opteron 248, 2.2GHz, 1MB cache, 8GB memory, RHEL4
ls1_tsh> uname -a
Linux ls1.lmb.internal 2.6.9-22.0.1.ELsmp #1 SMP Tue Oct 18 18:39:02 EDT 2005 x86_64
x86_64 x86_64 GNU/Linux
ls1_tsh> grep -V
grep (GNU grep) 2.5.1


grepping thro an 800 MB file:

ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
41.618u 1.263s 0:43.30 99.0%    0+0k 0+0io 0pf+0w
ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
41.675u 1.477s 0:43.54 99.0%    0+0k 0+0io 0pf+0w

ls1_tsh> ls -l  /tshgrp0/tsh/junkjunk
-rw-r--r--  1 tsh tsh 869243540 Sep  5 13:13 /tshgrp0/tsh/junkjunk

I'm certain the file is all stored in cache during this, and
that no disk accesses are taking place, because:
1. 'top' shows grep at 99.9% cpu
2. iostat on the disk in question shows zero activity.

The P4 box:
-----------

Monocore P4, 3.4GHz, 1MB cache, 2GB memory, FC3.
fm22_tsh> uname -a
Linux fm22.lmb.internal 2.6.11-1.35_FC3 #1 Mon Jun 13 00:52:08 EDT 2005 i686 i686 i386
GNU/Linux
fm22_tsh> grep -V
grep (GNU grep) 2.5.1


fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
0.272u 0.318s 0:00.59 98.3%     0+0k 0+0io 0pf+0w
fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
0.311u 0.281s 0:00.59 100.0%    0+0k 0+0io 0pf+0w

fm22_tsh> ls -l /scr0/tsh/junkjunk
-rw-r--r--  1 tsh users 869243540 Sep  5 14:29 /scr0/tsh/junkjunk


This is indicating that grep on the Opteron is *forty* times
slower than on the P4. Any ideas why this might be?
64-bit nasties in grep? Has anyone else seen this?

Cheers,
Terry

[Jeroen van Meeuwen]

T. Horsnell wrote:
>  I'm seeing huge discrepancies in grep speed between
>  my Opteron box and P4 boxes.
>  
>  The Opteron box:
>  ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
>  41.618u 1.263s 0:43.30 99.0%    0+0k 0+0io 0pf+0w
>  
>  The P4 box:
>  fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
>  0.272u 0.318s 0:00.59 98.3%     0+0k 0+0io 0pf+0w 
Isn't 0.272u a lot faster then 41.618u? Are these the real, user or
system times?

Kind regards,

-kanarip

-- 
Jeroen van Meeuwen, LPIC-1, MCP, CCNA
http://www.kanarip.com/
C6B0 7FB4 43E6 CDDA D258  F70B 28DE 9FDA 9342 BF08

[Terry Horsnell]

-- Start of PGP signed section.
[Charset ISO-8859-1 unsupported, filtering to ASCII...]
> T. Horsnell wrote:
> > I'm seeing huge discrepancies in grep speed between
> > my Opteron box and P4 boxes.
> > 
> > The Opteron box:
> > ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
> > 41.618u 1.263s 0:43.30 99.0%    0+0k 0+0io 0pf+0w
> > 
> > The P4 box:
> > fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
> > 0.272u 0.318s 0:00.59 98.3%     0+0k 0+0io 0pf+0w
>
> Isn't 0.272u a lot faster then 41.618u? Are these the real, user or
> system times? 
It certainly is.
1st column (u) = time spent in user mode
2nd column (s) = time spent in system mode
3rd column     = total elapsed time.

The P4 box did it in 0.59 secs, the Opteron took 43.3 secs.

Cheers,
T.

> Kind regards,
>
> -kanarip
>
> -- 
> Jeroen van Meeuwen, LPIC-1, MCP, CCNA
> http://www.kanarip.com/
> C6B0 7FB4 43E6 CDDA D258  F70B 28DE 9FDA 9342 BF08
>  
-- End of PGP section, PGP failed!

> -- 
> fedora-list mailing list
> fedora-list(a)redhat.com
> To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list

[Tim Waugh]

You're comparing different grep package releases (RHEL 4 vs FC 3), with
different patches, so it's hard to know what to take from the test
results.

Try comparing the same grep packages (name-version-release) so that you
know the same patches have been applied on both machines.

Tim.
*/

[Matt Davey]

On Tue, 2006-09-05 at 16:04 +0100, Tim Waugh wrote:
>  You're comparing different grep package releases (RHEL 4 vs FC
> 3), with
>  different patches, so it's hard to know what to take from the test
>  results.
>  
>  Try comparing the same grep packages (name-version-release) so that you
>  know the same patches have been applied on both machines.
>  
>  Tim.
>  */ 
Check your LANG environment variable.

I found that with a utf locale (eg 'en_IE.UTF-8'), many grep patterns
became incredibly slow.  Setting my locale to 'C' fixed the issue.  I
didn't have time to investigate further.

Matt Davey		When I read about the evils of drinking, I gave 
mcdavey(a)mrao.cam.ac.uk 	     up reading.  - Henny Youngman.

[Terry Horsnell]

> On Tue, 2006-09-05 at 16:04 +0100, Tim Waugh wrote:
> > You're comparing different grep package releases (RHEL 4 vs FC 3), with
> > different patches, so it's hard to know what to take from the test
> > results.
> > 
> > Try comparing the same grep packages (name-version-release) so that you
> > know the same patches have been applied on both machines.
> > 
> > Tim.
> > */
>
> Check your LANG environment variable.
>
> I found that with a utf locale (eg 'en_IE.UTF-8'), many grep patterns
> became incredibly slow.  Setting my locale to 'C' fixed the issue.  I
> didn't have time to investigate further. 
I had up2date'd grep on the Opteron system before I saw this msg,
and the problem then disappeared, but it was also present on a
Centos box running the un-up2date'd grep, and disappeared when
I switched $LANG from en_GB.UTF-8 to C.

Looks like the update has fixed the locale-sensitivity problem.

Thanks everyone!

Cheers,
Terry.


> Matt Davey		When I read about the evils of drinking, I gave 
> mcdavey(a)mrao.cam.ac.uk 	     up reading.  - Henny Youngman.
>
> -- 
> fedora-list mailing list
> fedora-list(a)redhat.com
> To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
>

[Nigel Wade]

T. Horsnell wrote:
>  I'm seeing huge discrepancies in grep speed between
>  my Opteron box and P4 boxes.
>  
>  The Opteron box:
>  ----------------
>  
>  Dual Opteron 248, 2.2GHz, 1MB cache, 8GB memory, RHEL4
>  ls1_tsh> uname -a
>  Linux ls1.lmb.internal 2.6.9-22.0.1.ELsmp #1 SMP Tue Oct 18 18:39:02 EDT 2005 x86_64
> x86_64 x86_64 GNU/Linux
>  ls1_tsh> grep -V
>  grep (GNU grep) 2.5.1
>  
>  
>  grepping thro an 800 MB file:
>  
>  ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
>  41.618u 1.263s 0:43.30 99.0%    0+0k 0+0io 0pf+0w
>  ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
>  41.675u 1.477s 0:43.54 99.0%    0+0k 0+0io 0pf+0w
>  
>  ls1_tsh> ls -l  /tshgrp0/tsh/junkjunk
>  -rw-r--r--  1 tsh tsh 869243540 Sep  5 13:13 /tshgrp0/tsh/junkjunk
>  
>  I'm certain the file is all stored in cache during this, and
>  that no disk accesses are taking place, because:
>  1. 'top' shows grep at 99.9% cpu
>  2. iostat on the disk in question shows zero activity.
>  
>  The P4 box:
>  -----------
>  
>  Monocore P4, 3.4GHz, 1MB cache, 2GB memory, FC3.
>  fm22_tsh> uname -a
>  Linux fm22.lmb.internal 2.6.11-1.35_FC3 #1 Mon Jun 13 00:52:08 EDT 2005 i686 i686 i386
> GNU/Linux
>  fm22_tsh> grep -V
>  grep (GNU grep) 2.5.1
>  
>  
>  fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
>  0.272u 0.318s 0:00.59 98.3%     0+0k 0+0io 0pf+0w
>  fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
>  0.311u 0.281s 0:00.59 100.0%    0+0k 0+0io 0pf+0w
>  
>  fm22_tsh> ls -l /scr0/tsh/junkjunk
>  -rw-r--r--  1 tsh users 869243540 Sep  5 14:29 /scr0/tsh/junkjunk
>  
>  
>  This is indicating that grep on the Opteron is *forty* times
>  slower than on the P4. Any ideas why this might be?
>  64-bit nasties in grep? Has anyone else seen this?
>  
>  Cheers,
>  Terry
>   
I find it hard to believe that any proessor could really grep an 800MB file in 
1/4 of a second. I think your P4 system is lying to you.

For example, grepping a 600MB RHEL CD iso I get the following figures:

2.8GHz Pentium 4:
real    0m20.531s
user    0m17.138s
sys     0m0.934s

2.6GHz Opteron 852:
real    0m9.858s
user    0m9.415s
sys     0m0.412s

2.66GHz Xeon:
real    0m18.545s
user    0m17.559s
sys     0m0.970s

Are you sure that the P4 system is really looking at the same file, and is 
successfully grepping the entire contents? It doesn't look like it is to me.

-- 
Nigel Wade, System Administrator, Space Plasma Physics Group,
             University of Leicester, Leicester, LE1 7RH, UK
E-mail :    nmw(a)ion.le.ac.uk
Phone :     [phone], Fax : [phone]

[Terry Horsnell]

> T. Horsnell wrote:
> > I'm seeing huge discrepancies in grep speed between
> > my Opteron box and P4 boxes.
> > 
> > The Opteron box:
> > ----------------
> > 
> > Dual Opteron 248, 2.2GHz, 1MB cache, 8GB memory, RHEL4
> > ls1_tsh> uname -a
> > Linux ls1.lmb.internal 2.6.9-22.0.1.ELsmp #1 SMP Tue Oct 18 18:39:02 EDT 2005 x86_64
> x86_64 x86_64 GNU/Linux
> > ls1_tsh> grep -V
> > grep (GNU grep) 2.5.1
> > 
> > 
> > grepping thro an 800 MB file:
> > 
> > ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
> > 41.618u 1.263s 0:43.30 99.0%    0+0k 0+0io 0pf+0w
> > ls1_tsh> time grep rubbishrubbish /tshgrp0/tsh/junkjunk
> > 41.675u 1.477s 0:43.54 99.0%    0+0k 0+0io 0pf+0w
> > 
> > ls1_tsh> ls -l  /tshgrp0/tsh/junkjunk
> > -rw-r--r--  1 tsh tsh 869243540 Sep  5 13:13 /tshgrp0/tsh/junkjunk
> > 
> > I'm certain the file is all stored in cache during this, and
> > that no disk accesses are taking place, because:
> > 1. 'top' shows grep at 99.9% cpu
> > 2. iostat on the disk in question shows zero activity.
> > 
> > The P4 box:
> > -----------
> > 
> > Monocore P4, 3.4GHz, 1MB cache, 2GB memory, FC3.
> > fm22_tsh> uname -a
> > Linux fm22.lmb.internal 2.6.11-1.35_FC3 #1 Mon Jun 13 00:52:08 EDT 2005 i686 i686
> i386 GNU/Linux
> > fm22_tsh> grep -V
> > grep (GNU grep) 2.5.1
> > 
> > 
> > fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
> > 0.272u 0.318s 0:00.59 98.3%     0+0k 0+0io 0pf+0w
> > fm22_tsh> time grep rubbishrubbish /scr0/tsh/junkjunk
> > 0.311u 0.281s 0:00.59 100.0%    0+0k 0+0io 0pf+0w
> > 
> > fm22_tsh> ls -l /scr0/tsh/junkjunk
> > -rw-r--r--  1 tsh users 869243540 Sep  5 14:29 /scr0/tsh/junkjunk
> > 
> > 
> > This is indicating that grep on the Opteron is *forty* times
> > slower than on the P4. Any ideas why this might be?
> > 64-bit nasties in grep? Has anyone else seen this?
> > 
> > Cheers,
> > Terry
> > 
>
> I find it hard to believe that any proessor could really grep an 800MB file in 
> 1/4 of a second. I think your P4 system is lying to you.
>
> For example, grepping a 600MB RHEL CD iso I get the following figures:
>
> 2.8GHz Pentium 4:
> real    0m20.531s
> user    0m17.138s
> sys     0m0.934s
>
> 2.6GHz Opteron 852:
> real    0m9.858s
> user    0m9.415s
> sys     0m0.412s
>
> 2.66GHz Xeon:
> real    0m18.545s
> user    0m17.559s
> sys     0m0.970s
>
> Are you sure that the P4 system is really looking at the same file, and is 
> successfully grepping the entire contents? It doesn't look like it is to me. 
Yes, quite sure. The P4 box has 2GB mem and the whole file fits
into the buffer cache. So after an initial slow grep to get the
file into memory, the rest are that fast.

I've up2date'd the grep utility on the Opteron box, and it now does the
grep in 0.9 secs! (again, this is an 8GB mem machine, so the whole file
will fit into buffer cache).

Cheers,
Terry

> -- 
> Nigel Wade, System Administrator, Space Plasma Physics Group,
>              University of Leicester, Leicester, LE1 7RH, UK
> E-mail :    nmw(a)ion.le.ac.uk
> Phone :     [phone], Fax : [phone]
>
> -- 
> fedora-list mailing list
> fedora-list(a)redhat.com
> To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
>