> Is there any pointers out there about how to use the mysql + M5 integration?
Unfortunately, I'm pretty sure that there's nothing much. Basically,
the output side is easy, you just have to call m5.stats.initMySQL()
and then when stats are dumped, they get dumped to the mysql database.
There are two other issues. Database setup is done with
"util/stats/stats.py database init". The main issue is with
extraction.
Look at the stats.py script for example code. There are a few things
that I support:
First, once you've connected to the database, the code queries the
database for all stat names and sucks them out of the database. You
can then use the update_dict call to put all of the stats into a dict
of your choice (you can even pass globals() or locals() as the dict).
In that dict, you will get a whole bunch of proxy objects, for each
stat. So if you have an object foo.bar and a stat called blah, you'd
end up with foo.bar.blah in that dict. The proxies understand all of
the fomula stuff, so you can say
x = foo.bar.blah / foo.bar.baz and you'll get a new proxy object x
that is a division of the two stats.
You can call db.get("my.job.name", x) to query the database for the result.
There are a few more sophisticated things you can do. If you dump
stats more than once in a single run, you can use: db.sum(), db.avg(),
db.stdev(). Those stats give you the sum, mean, and standard
deviation of all of the dumps done in that single run.
All in all, it's really pretty powerful, but it does take some getting
used to. There are some outstanding changes that you'd probably find
useful in our shared patch queue (see stats.diff). That patch also
has lots of examples of how I integrated the database stuff with
matplotlib.
I have grand plans to go a lot further with this stuff. Essentially,
I plan to create a database file (probably a python shelf, though
maybe sqlite db) that contains all of the stat data and make all of
this python code both able to query the mysql database and the
database file. I will move all of the stats output code that is
currently in C++ over to python (I've made stats accessible to the
python side of M5). Then I can use the same basic query code to dump
the text file from either the mysqldb, database file, or the python
interface to stats. Finally, I will probably also move from MySQLdb
to sqlalchemy.
Ok, long e-mail. I hope you find this stuff useful. If you want to
learn a thing or two about python, databases, swig, and the C/Python
API, consider helping me out! This sounds like a whole lot of stuff,
but in all honestly, a significant fraction of the code is written.
It's mostly a matter of putting the pieces together and cleaning it
all up.
Nate
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
