Re: [bug-gawk] gawk 3.1.7 on RHEL6 much slower than 3.1.3 on RHEL 4.7?

[Joseph Watkins]

On Sun, Oct 28, 2012 at 9:35 AM, Aharon Robbins <address@hidden> wrote:
> Hi.  Thanks for this:
>
>> Date: Tue, 16 Oct 2012 08:58:33 +0100
>> From: Joseph Watkins <address@hidden>
>> To: address@hidden
>> Subject: [bug-gawk] gawk 3.1.7 on RHEL6 much slower than 3.1.3 on RHEL 4.7?
>>
>> Hi all,
>>
>> I have two identical servers in terms of hardware (actually vm's), one
>> at RHEL 4.7 gawk 3.1.3, one at RHEL 6 gawk 3.1.7
>>
>> On RHEL6 gawk is very much slower. I did a simple test on a  3.7M line
>> text file, with a simple gawk command:
>>
>> On RH4.7, gawk 3.1.3 I get:
>>
>> time gawk '/^pat1/&&/pat2/ {x++};END {print "x="x}' file
>> x=40874
>>
>> real 0m7.266s user 0m6.809s sys 0m0.460s
>>
>> On RH6, gawk 3.1.7 I get same x result, with times:
>>
>> real 1m28.138s user 1m27.657s sys 0m0.453s
>>
>> 7s vs 1m28s! Initially I thought something had happened to the vm
>> after upgrade, but other tests of CPU (e.g. gzips, the same program in
>> perl) produce similar results on each. So it seems to be down to gawk.
>> Not sure this is the correct place to ask, but are there known
>> performance problems with gawk 3.1.7 on RH 6?
>>
>> Cheers,
>> Joe
>
> As Eli Zaretskii answered, it is likely related to locales. If you do
>
>         export LC_ALL=C
>
> and then run your tests, you should see 3.1.7 giving comparable speeds
> to 3.1.3, which dates from before locale support was added to gawk.
>
> FYI, 3.1.7 is 3 years old: Redhat should upgrade. The last released
> 3.1 version was 3.1.8, and there are a few fixes over that in the git
> repo.  However, gawk 4.0.1 is the latest official released version which
> has many feature improvements and some bug fixes.  You may wish to try
> that version. I think that even without LC_ALL=C for basic things it
> is just as (or almost just as) fast as 3.1.3.
>
> Hope this helps,
>
> Arnold

Hi, I think my reply to Eli didn't go to the list, so to confirm:

Well, your guess is correct! The 4.7 server was set to C, but the 6
one was UTF-8
Changing to C makes it go about 9 times faster (10s vs 90s).
That's a *big* difference - will certainly remember that one.
Thanks very much.

Thanks Arnold for the further thoughts. I'll see what our sysops will
allow in terms of upgrades.

Cheers