On Sat, Aug 11, 2012 at 8:13 PM, Jon Chiappetta
Ok so my quest for more reasonable build time comparisons is still
must be really bad at this... So I tried two new versions of the script, one
Don't be hard on yourself. I don't think this task is easy.
Unfortunately for you (and me) I have the horrible ability to look at
koji build stuff and know pretty closely whether it looks sane or not.
This isn't your help, I just need to go and do something else :-P
At times as a programmer it's hard to see the forrest for the trees.
We'll get there.
which just took PA's longest buildarch time and SA's shortest
time, however, the overall times summed up turned out to be way too
favorable on our behalf. The latest version now sums up the buildarch times
and divides them by how many of them there were. I think its a bit better
because the faster builders with faster build times will pull up the slower
build times, essentially its just an average calculation. The offset now for
us looks like around 15 days (half a month slower)? Anyway, the page is the
Interesting, that's currently probably about right with the current
set of builders. When we go to just yosemite/rainer builders I think
this will close out quite a bit which is why I'd be interested to see
in particular the guru/smarttop devices removed (and even the
cdot-panda* due to the nfs loopback because they're slow compared to
the hsv-panda* with the local usb disk).
same. I still haven't filtered out our slowest builders from this
either so the number of days could probably go down if I did. Let me know if
these numbers seem a bit more reasonable!
Looking good. It's interesting as it brings up some statistical
wierdness even on mainline  so it just goes to show it's not just
arm that has random builds :-)