Re: [Chicken-users] entry/exit to Scheme really slow on Solaris

[Bruce Hoult]

On Mon, 2003-08-11 at 15:23, Bruce Hoult wrote:
> This turns out to be a significant improvement not only for very short
> calls into Scheme, but also for normal programs.  For example, it
> dramatically lowers all the times for nsample, especially with small
> nursery sizes:

It helps other platforms, too, though to a smaller extent:

Athlon 1800+, Linux:

Before:
checking for the best nursery size ...
 8K -> 219 ms
 10K -> 200 ms
 12K -> 189 ms
 16K -> 172 ms
 24K -> 155 ms
 32K -> 148 ms
 48K -> 141 ms
 64K -> 139 ms
 128K -> 162 ms
 256K -> 253 ms
 512K -> 677 ms
 1024K -> 690 ms

After:
checking for the best nursery size ...
 8K -> 198 ms
 10K -> 183 ms
 12K -> 175 ms
 16K -> 163 ms
 24K -> 149 ms
 32K -> 142 ms
 48K -> 142 ms
 64K -> 136 ms
 128K -> 163 ms
 256K -> 215 ms
 512K -> 677 ms
 1024K -> 687 ms


HPPA 9000/785, HP/UX 10.20

Before:
checking for the best nursery size ...
 8K -> 2266 ms
 10K -> 1993 ms
 12K -> 1810 ms
 16K -> 1570 ms
 24K -> 1323 ms
 32K -> 1200 ms
 48K -> 1070 ms
 64K -> 1020 ms
 128K -> 903 ms
 256K -> 843 ms
 512K -> 840 ms
 1024K -> 913 ms

After:
checking for the best nursery size ...
 8K -> 1770 ms
 10K -> 1590 ms
 12K -> 1480 ms
 16K -> 1320 ms
 24K -> 1160 ms
 32K -> 1080 ms
 48K -> 990 ms
 64K -> 960 ms
 128K -> 870 ms
 256K -> 830 ms
 512K -> 830 ms
 1024K -> 903 ms


Updated times for my minimal call to Scheme (to add two numbers):

                                Before     After
Athlon 1800+, RedHat 8 Linux  :  2 sec    1.2 sec
HPPA 9000/785, HP/UX 10.20    : 29 sec   15.7 sec 
Sunfire v120, Solaris 9       : 94 sec    8.9 sec

So while not as important as on Solaris, this patch can nearly double
speed on Linux and HP/UX when you have extremely small and frequent
minor GCs.

-- 
Bruce Hoult <address@hidden>