On Sun, May 8, 2011 at 5:07 AM, Mariano Martinez Peck <marianopeck@gmail.com> wrote:


What approach do you suggest should I take?

I have no idea. �You could try and break down the various activities and benchmark them separately I suppose. �You could look at the profiles you're already generating in higher resolution to see where the time is going. �You could also try and extract the SystemProfiler from OpenQwaq, which will run on Cog, and is better than MessageTally. �Andreas wrote it to get around some of MessageTally's limitations. �It should hopefully show primitives quite accurately.


Hi. I have already read this several times. From you now, from Alexandre Bergel, from Henrik Johansen, etc. But no one (or maybe I didn't see it) give a list of the real existent MessageTally limitations or problems. So far, the only thing I read is what Eliot says here: "show primitives quite accurately"

So...which are the rest and what is solved by the SystemProfiler from OpenQwaq ??

MessageTally works by using delays to cause a high-priority process to interrupt the process being profiled. �So the sample is taken only at a suspension point. �The SystemProfiler support in the VM takes a sample either after a primitive has run or when the VM is checking for events, so it takes more accurate samples. �Further, the SystemProfiler's VM support also samples the active process when collecting the sampled method, so it more accurately ascribes profiles to processes. �MessageTally only profiles a single process.

�Both suffer from a bug my colleague Yaron Kashai observed this week at Cadence, which is that if GC pauses are long then both profilers will under-measure the GC costs because they both take a single sample before measurement. �So e.g. if a GC pause lasts five times longer than the poll time only one sample will be takem. �Ideally the SystemProfiler would be extended to maintain a list of sampled methods and processes.
�

best,
Eliot


Thanks in advance,


Mariano



�
HTH
Eliot
�

Cheers
Philippe






--
Mariano
http://marianopeck.wordpress.com