As I was driving home from Notre Dame yesterday, I drove south into a snow storm, which is really odd. Normally, it's the other way around -- you go North to get snow.
Had a good couplea days at ND.
I had a rockin' LAM pow-wow with Arun, Brian, Ron, and Dog. Dog was more of an observer, but he has been an official Friend of LAM for quite some time. When he gets some Spare Time(TM), he does need a Master's project, so it's possible that he'll do something in LAM. We'll see.
We discussed all the things that are Going On in LAM, and came to a few decisions:
- The next release of LAM will be 6.5, not 6.3.3. Mainly PR reasons, but also to signify that this is quite a big change since [the currently available] 6.3.2.
- First order of business this semester is to get 6.5 out the door. There's one or two issues that I'm going to look into this weekend, and then start giving tarballs to Ron and Brian for formal testing.
- Ron will probably start looking into Totalview support. That will be way cool; having a real parallel debugger that supports LAM.
- Brian is going to start looking into IPv6 support. This could give us some really cool things, such as optimized collectives (using IPv6's native multicast ability), security in the lamd (using IPsec), etc.
- Arun's going to finish the Myrinet RPI. He's having problems with long messages right now; hopefully that will get fixed Real Soon Now. He'll likely look into the VIA RPI after that, and dabble a bit in compression at the RPI level. This is an interesting sub-note: I think I had the inspiration to use compression in MPI during a drive SBN<-->Louisville. Sometimes it's not worth it, but sometimes it may make a huge difference in terms of bandwidth. It would be our ringer for ping-pong tests. :-)
- We'll probably have a series of quicker sub-releases (hopefully!) that incorporate major new features. e.g., 6.5.1 may have Myrinet support. 6.5.2 may have Totalview support. 6.5.3 may have some TCP RPI optimizations (e.g., tiny messages, fixed linked list handling). And so on. We can't really do this now because the 6.5 tree is very different than the 6.3.2 tree.
Didn't get to see Ed-n-Suzanne too much; maybe we'll have to do dinner one of these times when I go up there. Cleo went barking crazy when I came home both nights. I think the Cleo's non-barking acceptance rate is complicated function. There are multiple factors:
- Whether I initially come in during the day or at night (
day, 1 = day, 0 = night) - Whether Cleo is there when I initially arrive (only if during the day) (
at_home, 1 = home, 0 = not home) - Whether I come home at night by car or by foot (
car, 1 = by car, 0 = by foot) - How many days I have been there (
days)
- Phase of the moon (
moon, fraction from 0 to 1)
These factors have led me to the following equation (too bad mathML isn't yet implemented anywhere...):
chance_of_bark_at_night = | \sum{i = 0}{1 \step{.1}}{\frac{1}{days} \times ((day * .75) (at_home * .75))^{(i != moon)}} |
A team of 13 scientists that have been studying my visits to Chez Costech came up with that formula. I'm quite sure it's right.