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
On Fri, Oct 21, 2011 at 7:12 PM, Paul DeBruicker <pdebruic@gmail.com> 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.
Thanks
Paul
-- Mariano http://marianopeck.wordpress.com
On 21 October 2011 21:24, Mariano Martinez Peck <marianopeck@gmail.com> wrote:
On Fri, Oct 21, 2011 at 7:12 PM, Paul DeBruicker <pdebruic@gmail.com> 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.
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.
participants (4)
-
Igor Stasenko -
Mariano Martinez Peck -
Paul DeBruicker -
Stéphane Ducasse