I'm currently porting the Pharo VM to Solaris 11 for a specific project (Sparc and x86). I was wondering how many users prefer the stack VM, as for my own purposes the JIT is better (better concurrency when not using OS functions on a 128 thread machine, which is the deployment target). However since it wouldn't be that much extra work to port the stack VM, if there is interest I will do that as well.
thanks
Andrew Glynn
|
hi Andrew,
this is soooooo good! will it be published anytime soon? and as an answer, nobody wants the StackVM… is just there as a backdoor if the JIT does not works, but if you manage to compile the cog version, stack would be trivial (and is always good to have a ci job that manages all that). Esteban
|
In reply to this post by aglynn42
-- Does this mail seem too brief? Sorry for that, I don’t mean to be rude! Please see http://emailcharter.org . Johan Fabry - http://pleiad.cl/~jfabry PLEIAD and RyCh labs - Computer Science Department (DCC) - University of Chile
|
In reply to this post by aglynn42
On Thu, Jun 2, 2016 at 10:08 PM, aglynn42 <[hidden email]> wrote:
> I'm currently porting the Pharo VM to Solaris 11 for a specific project > (Sparc and x86). I was wondering how many users prefer the stack VM, On a platform with a JIT VM, no one uses the Stack VM. But as I understand it, the reduced complexity of the StackVM as a pure interpreter is easier to port and a pragmatic stepping stone. Perhaps its not so critical since the cpu target is the same. One advantage of having both is helping to isolate problems to the JIT or non-JIT code. btw, I recently started playing with the VM Simulator and found it easier to follow execution of the StackVM to learn VM architecture, which will help highlight JIT differences when I soon try the JIT simulator. Here is some info if you want to try it... http://forum.world.st/Exploring-the-simulator-was-Re-REPL-image-for-simulation-td4896870.html#a4898186 You might like to subscribe to the VM mail-lists... http://lists.squeakfoundation.org/mailman/listinfo/vm-dev http://lists.squeakfoundation.org/mailman/listinfo/vm-beginners cheers -ben > as for > my own purposes the JIT is better (better concurrency when not using OS > functions on a 128 thread machine, which is the deployment target). However > since it wouldn't be that much extra work to port the stack VM, if there is > interest I will do that as well. thanks Andrew Glynn |
When you say you are porting the Pharo VM for Solaris 11 and Sparc/x86, you mean you are adding support for the Sparc ABI for FFI calls and a Sparc back-end to the JIT ?
On Thu, Jun 2, 2016 at 6:29 PM, Ben Coman <[hidden email]> wrote: On Thu, Jun 2, 2016 at 10:08 PM, aglynn42 <[hidden email]> wrote: |
Sorry for the delay in answering - I've been buried in this project.
That is the idea, though as Ben pointed out, not using Cogit initially particularly when moving to Sparc, reduces the complexity, so one step at a time. The x86 version with Cogit is nearly debugged and will initially be used for a project where I have written code (already tested on Oracle Linux 7.2) to monitor BGP events on Cisco aggregated service routers (ASR9010's) to track dynamic topology changes. We're deploying the first version on surplus Sun x4140 12 core / 24 thread x86 servers, and the later Sparc version to handle more routers on surplus T5120 / T5220 machines. The reason for the use of Solaris and surplus Sun machines is a combination of keeping the cost down while supporting the throughput rates - ASR9010's generate an average of 57m BGP events / minute, and the API is straight XML. We eventually need to monitor ~200 ASR's. Once the code is stable I will create a sub-project for the Solaris code on Github. thanks ANdrew Glynn
|
Free forum by Nabble | Edit this page |