Click on SMP?

[Benjie Chen]

Hi Maurice,

Thanks for your interests in Click. Robert Morris and I have been working
on a Click SMP implementation. Our paper on Click SMP was accepted to this
year's Usenix conference.

Click SMP has not been merged into the main Click distribution yet, but 
you can download a tar file from the following location if you'd like to 
take a look at what we have done:

http://www.pdos.lcs.mit.edu/~benjie/click-smp.tar.gz

We still plan to make improvements and modifications to our implementation. 
If you have suggestions or bug fixes, please let us know. Furthermore, a 
draft of our Usenix paper can be found at:

http://www.pdos.lcs.mit.edu/~benjie/papers/usenix01.draft.ps

We would also welcome any comments on the paper as well. Please do not 
distribute the Click SMP tar file or the paper. Thanks.

Benjie



On Thu, Feb 01, 2001 at 03:58:08PM -0500, Marks, Maurice wrote:
> Hi Clickers! I'm working on a project at Compaq that has to do with routers
> and I've downloaded your code and tried it out. I'm very impressed with the
> concepts and the code. Since I'm working on Alpha rather than x86 I've just
> started making a few 64-bit fixes to get it to run on that platform. I'll
> send you the changes when I'm done. 
> 
> Anyway, I have a more significant question. I'm exploring ideas about
> router-like software running on SMP (or actually large NUMA-like)
> architectures. Have you thought about how to schedule multiple click
> processes or threads on that sort of thing? I'd like to be able to assign
> several processors to one routing configuration - where they would share the
> routing load, and perhaps have others executing other routing
> configurations. If I can make that work I would be able to combine multiple
> network infrastructure functions onto a pool of processors, possibly even
> doing some dynamic load balancing (assigning more or less processors to a
> particular routing configuration).
> 
> I dont know if you are interested in this sort of application but if you are
> let me know. What I'm going to do after I get the uniprocessor stuffing
> running well is to start adapting it for a SMP. This will probably involve
> surgery related to the scheduling and the interface to the IO adapters at a
> minimum. I'll make the results avilable to you.
> 
> So thanks for your work and publications. I'm finding it very useful.
> 
> best regards
> /maurice marks (senior fellow at Compaq)
>  

-- 
benjie chen
benjie at lcs.mit.edu