|
Paul DeBruicker |
|
|
Hi -
When should the multi-threaded version of Cog be used vs the single threaded version? Is either one more stable or performant, etc? Thanks Paul |
|
Mariano Martinez Peck |
|
|
On Fri, Oct 21, 2011 at 7:12 PM, Paul DeBruicker <[hidden email]> wrote: Hi - The MT of Cog is only for FFI as much as I understand. The VM itself continues to be a green thread, even with CogMT. So no, there is not performance change, only in FFI. That being said, I never hear someone apart from Eliot successfully using CogMT with real multithreaded FFI. There may be some work still to do before you can use it. Thanks -- Mariano http://marianopeck.wordpress.com |
|
Igor Stasenko |
|
|
On 21 October 2011 21:24, Mariano Martinez Peck <[hidden email]> wrote:
> > > On Fri, Oct 21, 2011 at 7:12 PM, Paul DeBruicker <[hidden email]> wrote: >> >> Hi - >> >> When should the multi-threaded version of Cog be used vs the single >> threaded version? Is either one more stable or performant, etc? >> > > The MT of Cog is only for FFI as much as I understand. The VM itself > continues to be a green thread, even with CogMT. So no, there is not > performance change, only in FFI. > That being said, I never hear someone apart from Eliot successfully using > CogMT with real multithreaded FFI. There may be some work still to do before > you can use it. > I plan to change the VMs on Jenkins server to build using MT version of (co)interpreter. I misunderstood Eliot's statement once, when he said that MT version of VM is 5% slower, but then Eliot clarified that this slowdown related only to FFI, not the whole VM. -- Best regards, Igor Stasenko. |
|
Stéphane Ducasse |
|
|
Thanks igor
yes indeed this is important. Stef >>> >>> Hi - >>> >>> When should the multi-threaded version of Cog be used vs the single >>> threaded version? Is either one more stable or performant, etc? >>> >> >> The MT of Cog is only for FFI as much as I understand. The VM itself >> continues to be a green thread, even with CogMT. So no, there is not >> performance change, only in FFI. >> That being said, I never hear someone apart from Eliot successfully using >> CogMT with real multithreaded FFI. There may be some work still to do before >> you can use it. >> > > I plan to change the VMs on Jenkins server to build using MT version > of (co)interpreter. > > I misunderstood Eliot's statement once, when he said that MT version > of VM is 5% slower, > but then Eliot clarified that this slowdown related only to FFI, not > the whole VM. > > > -- > Best regards, > Igor Stasenko. > |
| Powered by Nabble | See how NAML generates this page |