« Kenau Reeves can't act | Main | I'm the original Nicks Superfan »

Jeff's Journal

Strange things are afoot at the Circle K.

In the slashdot post that I mentioned in a prior jjc post, Pete actually magically got moderator privileges that day (all the moons must have been in alignment), and so he moderated my post up to a 5 (the highest possible value). LOL!! Like I said, my post used bad grammar and was slightly unclear. I find it extremely amusing that it got moderated all the way up to a 4 (without help from friends), and then finally pushed up to a 5 (with a little help :-).

For those of you who are wondering, here's my post.

Good quote from Brian's journal:

"Friends don't let friends integrate ROMIO with LAM"

I picked up Janna from the airport the other day; they had just returned from a holiday of hiking through Switzerland. They had a great time and did much walking throughout various out-of-the-way little Swiss towns.

They bought me a swiss army knife with my initials engraved on it, and got some chocolate for Tracy.

We ran into Aimee in the Louisville airport; she was on her way to some business meeting the next day. It's a small world.

Eric fixed the latency problems on the Babel cluster. It seems that the NICs were continually going into auto-negotiation mode (to swap between 10 and 100Mbps) for some reason, which caused all kinds of retransmission delays and errors. But now the cluster seems to be working well (it's amazing that it has been that way for at least a year) -- the NFS delays seem to be much more normal, meaning that the latency is about what you would expect from NFS.

I drove back from Bloomies yesterday. Had a good 2 days there; I'll typically be spending 1-2 days a week there.

It's very easy to get from Bloomies to my house -- it's essentially 3 highways: IN46 to I65 to I64.

While I was driving home yesterday, I was driving along the forest-lined IN46 when suddenly 3 C-130's flew above the tree tops right in front of me (military transport planes). They were flying North. They looked to be at about the right altitude for parachute jumping. IN46 is nowhere near any military bases that I know of. Weird.

Later, I got on to I65 and headed towards Louisville. About 40 miles south, I saw 3 C-130s again (must have been the same ones, although they were much higher) flying east as they flew over me on I65. About 20 miles further south, I saw them again, this time on the West side of I65, but flying due south.

The Louisville airport was somewhat nearby, and they were heading in that general direction, but Ft. Knox is an additional 75-100 miles further south; they might have been heading there as well. Both airports can handle C-130s. Who knows.

The only obvious conclusion that I can draw from this experience is that the government is shadowing me, and/or bombarding me with electro-magneto kinetic rays in attempts to steal my brain. "This line is tapped, so I must be brief."

If I disappear and/or turn into a Microsoft neophyte, you all will know the reason why. Let the truth be known; trust no one.

I came across a great term yesterday: "war driving". It's a moniker off the old term "war dialing" from back in the 70s and 80s. War driving is taking a laptop with a wireless NIC and literally driving around and seeing what wireless networks you can tap into.

There's at least one windoze-based utility program for this: http://www.personaltelco.net/index.cgi/NetStumbler.

How secure is your network?

The SSI project in LAM is going quite well; I'm pretty excited about it. More and more ideas about what it can be used for keep occurring to me; SSI may solve a lot of issues and provide a really nice framework inside of LAM. If we do it Right, it may end up being the One True Way that we integrate LAM to all new kinds of systems (PBS/TM, Grid, Scyld, KLAT, etc.). Indeed, I'd really like to be able to use SSI to integrate new algorithms into LAM -- such as the tree-based lamboot. Very cool stuff.

I chatted with Kay at IU yesterday; she's a "pre-faculty" in the CS department (analogous to how I'm a pre-post-doc). Her advisor is Andrew C. (formerly of UIUC); she's part of the Fast Messages Group. We had an interesting chat about Myrinet, research, software engineering, Windoze development, etc. She's somehow connected to our IPCRES group, but indirectly. I don't understand the exact relationship (I don't think anyone does, yet :-). But she'll be at least somehow connected to our group.

It seems that one of the big things that Fast Messages did to make Myrinet message passing fast was to support essentially the same thing that writev() does (although they call it "scatter / gather", which confused me until I realized that she wasn't talking about collectives) -- take pointers to different chunks of memory and write them all out into a single transfer, rather than:

  • Force the user to copy everything into a single, contiguous buffer so that it can be sent in a single transfer, or

  • Force the user to use multiple transfers to send everything.

GM doesn't currently have such a vectored-write (or vectored-read) capability. So I pinged the folks at Myri about it and asked if they will ever support such a thing. Indeed, that would help us in LAM --
it would effectively eliminate the need for the tiny message protocol in the gm RPI.

Post a comment

(If you haven't left a comment here before, you may need to be approved by the site owner before your comment will appear. Until then, it won't appear on the entry. Thanks for waiting.)


This page contains a single entry from the blog posted on July 12, 2001 9:59 AM.

The previous post in this blog was Kenau Reeves can't act.

The next post in this blog is I'm the original Nicks Superfan.

Many more can be found on the main index page or by looking through the archives.

Powered by
Movable Type 3.34