gcl-devel
[Top][All Lists]
Advanced

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

[Gcl-devel] Re: more random noise on timing


From: Camm Maguire
Subject: [Gcl-devel] Re: more random noise on timing
Date: 02 Mar 2006 18:59:48 -0500
User-agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2

Robert Boyer <address@hidden> writes:

> Using Camm's cvs suggestion to get a Dec 14 2.7.0 GCL, things seem unchanged
> between then and now in terms of the Nqthm proofs.  These are both with
> default MAXPAGES, ANSI.
> 
>    cvs -z9 -q co -d gcl -D 20051214 gcl
>    run-gbc time    :   1975.380 secs
>    child run time  :    147.250 secs
>    gbc time        :     59.520 secs
> 
>    Tuesday 2.7.0
>    run-gbc time    :   2000.970 secs
>    child run time  :    147.340 secs
>    gbc time        :     62.230 secs
> 
> This is good news, suggesting that rumors of a slowdown since December are
> entirely false.  Apologies for rumor mongering.  But it creates a real
> mystery for me.
> 
> Running in 2.7.0 just a few days ago, but in a huge (170 mb) saved image with
> very large ihs and vs stacks, I got 10% worse:
> 
>    run-gbc time    :   2181.560 secs
>    child run time  :    141.280 secs
>    gbc time        :     94.930 secs
> 
> So this supports Matt's position that size does matter, which I "disproved"
> yesterday, at least for a big MAXPAGE.  I'm a bit lost.  How could simply
> running in a saved image configured with with
> 
>             --enable-ihssize=2097152             \
>             --enable-vssize=2097152              \
>             --enable-maxpage=524288              \
>             --enable-holediv=100                 \
> 

My guess is holediv, though if so the timing output should be
improved.  Try holediv=16.  The default is 4, and you are using 4x the
default maxpages.

The gc time is up by 50%, supporting this hypothesis.   However, to be
material, this would have to mean that a substantial fraction of the
actual gc overhead is being accumulated in the run-gbc time statistic,
which at this point is just "idle speculation" :-).

Take care,

> cause a 10% slowdown?  Or is this all some sort of "operator error" on my
> part?  Sorry for wasting your time.
> 
> Bob
> 
> 
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah




reply via email to

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