Hi Avadh,

Let's say I wanted to port my stats collection over to the stats module you
guys are implementing. I started looking at the code in ptlsim/stats/ but
it's a bit unclear to me what I should be focusing on.

It seems that stats.h and dataStore.* belong to the old stats system --
should I just be ignoring these files and only looking at statsBuilder.* ?

Going forward, are you guys planning on supporting both BSON and YAML?

-Paul

On Thu, Jun 2, 2011 at 12:40 PM, avadh patel <[email protected]> wrote:

> This indeed seems very interesting and useful to many people.
> Please send me the patches or link to your git repo.
>
> About the new stats module we might find a way to incorporate it into new
> Statistics Builder module. It can be categorized as time/cycle relative
> counters that can be dumped into separate stats/csv file.  Also we can
> easily implement dumping new Statistics Builder counters to new format.
>
> One thing that I personally would get hands to is your gmail poll script
> that email back the response of running simulations.
>
> - Avadh
>
> On Wed, Jun 1, 2011 at 1:50 PM, DRAM Ninjas <[email protected]> wrote:
>
>> Hello All,
>>
>> While using marss I've developed a kind of ad hoc set of tools comprised
>> of various scripts that make my life easier. I've been considering packaging
>> them up and throwing them out to a github repo for people to use, but I
>> wanted to gauge interest before I took the time to clean these up and push
>> them out.
>>
>> Most of these are outside of MARSS with the exception of a stats module I
>> wrote. While the PTLStats is great for overall simulation stats, I don't
>> think it really has a time-based component (please correct me if I'm wrong).
>> The stats module I wrote is basically a single global object that acts like
>> a map where you can say
>>
>> mystats["bus_rejections"]++;
>> or
>> mystats["cache_latency"]+=5;
>>
>> It's nothing fancy -- just a single .h file with some definitions, but it
>> allows me to do averages, sums, per-cycle averages, etc. It automatically
>> dumps these stats to a file every X cycles and then a python script graphs
>> them. I've been reluctant to submit a patch for this since uh, it seems like
>> the last thing marss needs is another stats collection system. This one is
>> pretty lightweight and since it's mostly meant to be used with a script to
>> draw graphs, it doesn't have a lot of complexity of trying to be human
>> readable.
>>
>> The other scripts do things like:
>>
>> 1. Create a bunch of time varying plots (from the stats module described
>> above or any csv file you generate), throw them into a png. I've attached an
>> example png (not actually sure if the listserv allows that, but worth a
>> shot).
>> 2. Create some time-varying plots, throw them into an eps (which can be
>> post processed into pdf and imported as vector graphics into LaTeX).
>> 3. Log into a gmail account with an xoauth token (no password) and send an
>> email (with arbitrary) attachments (this one already exists in marss and can
>> be used with the -execute-after-kill patch I wrote a while ago)
>> 4. Display the simulation status (including cycle count, which cores are
>> in kernel/user/library RIPs)
>> 5. Poll a gmail account for a specific label -- when something appears in
>> that label, send back a reply with the status of all running simulation.
>>
>> So if any of this sounds interesting or useful, let me know and I'll
>> expend the effort to clean these things up and push them out somewhere.
>>
>> -Paul
>>
>> _______________________________________________
>> http://www.marss86.org
>> Marss86-Devel mailing list
>> [email protected]
>> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>>
>>
>
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel

Reply via email to