[Uim] Re: Massive slowdown after revision 129

[Etsushi Kato]

On Fri, Jan 07, 2005 at 08:13:37PM +0900,
YamaKen <yamaken at bp.iij4u.or.jp> wrote:

> > On 2005/01/07, at 15:15, YamaKen wrote:
> > 
> > > -      "-h16384:64",   /* heap_size(unit: lisp objects):nheaps */
> > > -      "-t16384",      /* heap_alloc_threshold (unit: lisp objects) */
> > > +       "-h100000:10",  /* heap_size(unit: lisp objects):nheaps */
> > > +      "-t100",        /* heap_alloc_threshold (unit: lisp objects) */
> > 
> > Yep.  Very slow key response with this siod setting in r148.
> > 
> > Changing heap_alloc_threshold actually affected response drastically.  
> > heap_size itself doesn't seems critical about this.
> 
> > With modified (slow) version,
> > 
> > [starting GC]
> > [GC took 0 cpu seconds, 1286 / 98714 cells collected in 10 / 10 heaps]
> 
> Is this really runs with "-h100000:10 -t100"? Very strange if so.

Arrg!  You are right.  I did use -h10000:10, sorry.

Now I tested -h100000:10 -t100 with r148, and it also showed massive
slowness.  Two to five GC took place with every key press.

[starting GC]
[GC took 0 cpu seconds, 2577 / 97423 cells collected in 1 / 10 heaps]

> This message indicates the interpreter allocates 11.5MB of heaps
> rather than 1.5MB of below. This heap size * GC frequency will
> directly impact performance.
> 
> > With normal (r148) version,
> > 
> > [starting GC]
> > [GC took 0 cpu seconds, 32161 / 98911 cells collected in 8 / 64 heaps]

I confirmed this test was correct.

> Please let me know additional information.
> 
> - VSZ and RSS of the two processes

I've conducted the test using leafpad with gtk-immodule.

normal one: VSZ 51652, RSS 17376
slow one: VSZ 51288, RSS 16976

Cheers,
-- 
Etsushi Kato
ekato at ees.hokudai.ac.jp