🚀 go-pugleaf

In article <JET.92Nov17165616@boxer.nas.nasa.gov> jet@boxer.nas.nasa.gov (J. Eric Townsend) writes:
>"patrick" == Patrick F. McGehearty <patrick@convex.COM> writes:
>JET == jet@boxer.nas.nasa.gov (J. Eric Townsend) 
>
>JET>Blinking lights.  Lots and lots of blinking lights....
>
>patrick> have suggestions for specific hardware events that might be
>patrick> connected to blinking lights on a massively parallel shared
>patrick> memory super?
>
>I sorta like the ipsc/delta/sigma solution:
> - one for cpu active
> - <something> for communication
> - one for node failure
>
>I've seen more than one person debug their code on the ipsc/860 by
>sitting in front of the cabinet and watching the led's.. "See, it hung
>in computation after the third massive data shuffle..."
>

Although I'm fond of lights, and have found uses for them, the information
they provide is limited to the machine room (unless you have a video link
to you workstation and a camera in front of the machine ;-)

I've implemented something like SGI's gr_osview into our locally developed
hypercube topology machine here.  It's much lower bandwidth than real lights,
but I can display cpu usage, message rates, byte communication rates, link
contention, etc. for each node in a matrix of bar graphs.  Click on the
graph and you get a detail including history of the last 30 secs of samples.

This has been quite helpfull towards giving me a feeling for how things are
progressing, but I'd really like a way to associate the events recorded in
the display with portions of the code.  I can see commiunications getting
congested but I don't know where in the code I might try to change things.

So while lights (or my remote bar graph mechanism) are good for seeing really
gross features of program behaviour, I'd like to find a mechanism to relate
such statistics to locations in the code causing them.  Does anyone know
of a system which can do this?  I'm a bit familiar with things like ParaGraph
events, but do any systems report low-level events in a way the programmer
can relate to the code?


-- 
Scott Townsend,  Sverdrup Technology Inc.  NASA Lewis Research Center Group
fsset@bach.lerc.nasa.gov