Tales of a Fourth Grade Nothing

I've just noticed that Chris Lazou's report on NUG for HPCwire is now available. I'm please to see a favourable mention of Professor Goede's fantastic talk on the beauty of CFD — easily my pick of the conference. In fact, the presentation was so amusing that, for days afterward, people were able to reduce me to a giggling mess merely by quoting a few choice phrases from it. Not something that could be said for most of the other speakers...

Couple of intersting sessions today. The site presentation from DKRZ was depressingly familiar. They mentioned that they'd seen a whole bunch of problems with Unitree and a whole bunch of crashes, the cause of which had yet to be diagnosed, and that GFS only really worked as long as nothing was changed. Had the delivery been only slightly livilier, I'd have been out of my seat, waving my arm above my head and shouting, "Testify, brother! Testify!"

Which more or less paved the way for Phil T's barnstorming speech where he stood up, kicked ass and took names. In a shocking lapse of the usual diplomacy, where you say how much you like the machine and then wheedle the conversation round to the point where you're bad mouthing stuff, he just came out and said what needed to be said: SUX is very dated and hasn't really changed since it was introduced in '90; that the compilers are substantially less than ideal and that the compiler group are more interested in a quiet life than actually improving their products. He also praised NEC Australia to the heavens and said that you apprechiated their job so much more when you saw all the crap they had to deal with from NEC Japan.

Actually, the best point from his talk — the one that didn't involve bad mouthing NEC — was the distinction between supercomputing and real time HPC. He said that supercomputing basically involved running one large job, whereas real time HPC involved running a complex job mix and having to worry about the precise start and end times of the jobs. Following his memetic onslaught, we all felt out mindsets shifting and went round calling ourselves real time HPC rather than supercomputing people. I'm actually one step ahead than some of the others — my business card already describes me as an HPC specialist.

Anyway, time to go and prepare for the gala dinner...

Today's sudden and unexpected news: I've been nominated to go to next month's conference in Toronto and as a result, I spent a frantic couple of hours trying to sort everything out before the registration deadline expired this afternoon. Having finally got all that sorted, I've now got to see if I can come up with paper for it...

Went to a few of the presentations today, although this morning was total lossage. Also discussed various hacks with Col and Saki - they reckon we can bump up the baud on the console to speed up the process of doing a bta in KDB when the system wigs.

Michael's presentation was pretty good, although there were a couple of moments when people asked tricky questions and he looked like a deer caught in the headlights. Richard's talk was interesting, but perhaps could have done with a couple of jokes - he also dropped me in it by talking up my (still unwritten) presentation and saying that it would probably cover scheduling. Hmmm, I must remember to put in some scheduling slides.

DKRZ site thingy was interesting - they archive between 3 and 15 terabytes of data per day, where as we struggle with one. Puts it all in perspective. They also mentioned a scheme to get funding a European Earth Simulator. I can't see that working - the Earth Simulator was basically a Keynesian boondoggle to allow the Japanese govt to dump cash into their high performance computing industry, kinda like the ASCI projects in the US - but since Europe doesn't have a huge HPC industry, it doesn't sound politic to me.

Thomas and Maryanne's talk on DMI was pretty interesting - they mentioned all the problems they'd had with ClusterPro (including an upgrade that trashed their operational forecast filesystem) and talked about their scheduling configuration. I'm looking forward tomorrow, when the High Commander has arranged a meeting with them to discuss common issues. I've also got to remember to ask about the hacks required to get NRPE working under SUX. Maybe if I put something here, I'll spot it tomorrow when I'm vanity surfing and it'll jog my memory.

The high point of the cocktails (ha! as if) and canapes evening was the following URL, which Robert kindly supplied, which converts regular web sites in Valley Girl Speak. Maybe I'll run it against my blog. On the downside, I was late leaving and I had to get the super slow round the hospital bus into town, which totally sucked, rather than the direct one which stopped running at like 7pm.

I've decided I'm going to go to bed early. I'm totally washed out with yesterday's late night and all - just what I needed on top of my recent program of aggressively sleep deprivation.