[gambit-list] parallel GC

Bradley Lucier lucier at math.purdue.edu
Mon Oct 31 16:35:47 EDT 2016


On 10/31/2016 04:25 PM, Marc Feeley wrote:
> The CPU time seems correct to me. With multiple threaded VMs the GC takes 25 seconds more CPU time than with single threaded VMs (where it takes 13 seconds).  This is probably because there is little parallelism that the GC can exploit (perhaps a small number of very large bignums?).


There are no bignums---there are a lot of Meroon objects representing a 
tringulation (triangles, edges, vertices, and the relationships between 
them).  About 15GB worth

> By the way, your screendump is strange… the middle part, the “percent profile”, should be a summation (put on a scale of 100%) of the activity bars above (one per processor).  But the red areas (for the GC) don’t line up…  Could you try to run xactlog with the -us or -ms option (to scale to microseconds or milliseconds)?

You mean like this?

heine:~/programs/gambit/gambit/contrib/xactlog> ./xactlog -us 
~/WWW/615-2016/software/gambit.actlog

It appears to be the same (eps file included).

Brad
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gambit.actlog.eps
Type: image/x-eps
Size: 398164 bytes
Desc: not available
Url : http://webmail.iro.umontreal.ca/pipermail/gambit-list/attachments/20161031/72fb209a/attachment-0001.bin 


More information about the Gambit-list mailing list