Post by votan on Sept 8, 2009 18:55:24 GMT -5
Carl, you are probably pretty much up to date about cincoms development progress.... and as I just read, that there will be soon VisualWorks7.7 coming (or is already out?) with multithreading/multicore support I wanted to ask if you know more about it?
I'm currently running RB on my testserver (HP Proliant DL380 G5) with 2xquadcore CPUs and it pretty much hurts to see RB making no real usage of that and slowing down pretty fast when a few simultanous requests come in...
Here is the original post:But way more informative might this link be: www.cincomsmalltalk.com/userblogs/arden/blogView
Check all topics under this link... many about multicore support..
Would be cool if you can tell more about it and maybe also about including it in a future RB version? Would solve many of the current performance issues. ....besides the fact that RB does not keep a compiled version of a project and always compiles it first upon initial request... but I guess that will already be fixed with the next release?
Hope I don't get flamed again.... RB is the greatest.. but performance needs to be improved. And multithreading/multicore support will provide a huge performance gain.
Add: Just read that the VW7.7 prerelease is out... so maybe you can take a look at it?.. but you probably already did...
Another add: Just experimented with CPU affinitiy on another small sever... and even though, The visualworks informations say, changing cpu affinitiy to just use one cpu would increase performance of the image by the factor of 2 because of avoiding cpu switching and extra chaching (on intel HT machines) it infact slows down my RB by the factor of two on an AMD machine.... will look into this more...
And another add.... tried changing cpu affinity on an intel core2duo t7400..... no speed changes at all... one core is as fast as 2 cores.... what means.. using default "use all cores" does not slow down RB.. but unfortunately it also does not speed up RB using all available cores. Seems like AMD did a more usefull implementation of their multicore systems.. but they are slower by default. Using both cores of my AMD X2 system is still about 20% slower than using just one core of my c2d T7400 system. But anyway, that does not help with using RB on my servers with 8 intel Xeon cores...
I'm currently running RB on my testserver (HP Proliant DL380 G5) with 2xquadcore CPUs and it pretty much hurts to see RB making no real usage of that and slowing down pretty fast when a few simultanous requests come in...
Here is the original post:
VW7.7
There will be a framework included, in preview, in VisualWorks 7.7 called
Polycephaly.
This is a framework for helping leverage multicore computers.
Its goals are a simplistic but effective use of multi cores.
I'll be speaking about this and some experiments I ran at ESUG.
Hth
There will be a framework included, in preview, in VisualWorks 7.7 called
Polycephaly.
This is a framework for helping leverage multicore computers.
Its goals are a simplistic but effective use of multi cores.
I'll be speaking about this and some experiments I ran at ESUG.
Hth
Check all topics under this link... many about multicore support..
Would be cool if you can tell more about it and maybe also about including it in a future RB version? Would solve many of the current performance issues. ....besides the fact that RB does not keep a compiled version of a project and always compiles it first upon initial request... but I guess that will already be fixed with the next release?
Hope I don't get flamed again.... RB is the greatest.. but performance needs to be improved. And multithreading/multicore support will provide a huge performance gain.
Add: Just read that the VW7.7 prerelease is out... so maybe you can take a look at it?.. but you probably already did...
Another add: Just experimented with CPU affinitiy on another small sever... and even though, The visualworks informations say, changing cpu affinitiy to just use one cpu would increase performance of the image by the factor of 2 because of avoiding cpu switching and extra chaching (on intel HT machines) it infact slows down my RB by the factor of two on an AMD machine.... will look into this more...
And another add.... tried changing cpu affinity on an intel core2duo t7400..... no speed changes at all... one core is as fast as 2 cores.... what means.. using default "use all cores" does not slow down RB.. but unfortunately it also does not speed up RB using all available cores. Seems like AMD did a more usefull implementation of their multicore systems.. but they are slower by default. Using both cores of my AMD X2 system is still about 20% slower than using just one core of my c2d T7400 system. But anyway, that does not help with using RB on my servers with 8 intel Xeon cores...