Supercomputing and Stuck in Canberra

[tcpip]

The past few days I've been attending the Challenges in High Performance Computing conference. It's a fairly small affair, and very much at the pointy-end of math nerds interested in such a subject. I managed to arrive in time for the conference dinner which was at the held at the Pollen Cafe at the Australian Botanical Gardens, which included a guided tour beforehand. The tour was really quite enjoyable and the food and drink were magnificent. Especially noteworthy presentations included Ulrich Ruede, from the University of Erlangen-Nuremberg, on "Extreme-Scale Resilient Multigrid Solvers", Paul Leopardi, from the Bureau of Meteorology, on "Optimizing workflow scheduling and capacity management", and Lois Curfman McInnes, from Argonne National Laboratory, on "Community Software Ecosystems". My own presentation was Why Computers Lie at Alarming Speed and the unum Promise. Also, after the second day, found myself at a Canberra Python group. After the conference went down to the National Computing Infrastructure to give a second presentation, this time an ARDC tech-talk on The International HPC Certification Forum and AU-NZ, which is the third time I've given that presentation in a month. Maybe there's interest in it or something. The NCI people gave a presentation on their new system, Gadi, which will be the peak system for Australia. Also was pleased to catch up with taavi for lunch where we chatted primarily about bronze-age economics.

Whilst these talks were going on, Tiger Airlines sent me a message that my flight back to Melbourne had been canceled due to engineering issues. You don't want to fly a Tiger that has engineering issues so I made my way back to the hotel to check-in for another two nights - the Nesuto - which provides essentially a small apartment for the price of a hotel room. It was quite comical in a way, it was pouring rain, I was dripping wet, and conversing incredulously on the mobile with Tiger about the fact that the next flight was in two night's time. The hotel receptionist could barely contain her laughter. Tiger will be paying for the extra two nights, along with a free flight home, which I suppose is not unreasonable under the circumstances. Tomorrow morning I'll visit the Legislative Assembly, given it's literally across the road from the hotel, the_shadow298 has arranged to catch up for lunch, and in the evening Zoe B., has invited me to a restaurant games night. So in a short time, I appear to be making something of a weekend of it after all.

Comments

[johnny9fingers]

One of my very old chums is a leading light in something called VDM. Do those initials mean anything to you? :)

[tcpip]

Vienna Development Method? That really is a narrow and pointy end of software. If you're in the business of building the formal methods for compilers, you're likely to attend world conferences with 20 people. :)

[johnny9fingers]

Yep.
He's at the Rutherford most of the time but has a few days a week in Town at Imperial. And the odd conference here and there.

I don't suppose it earns him a huge amount of money, but I reckon he's doing stuff that benefits society.

[tcpip]

I think so as well. We are so dependent on getting these intricate little components right, and we only notice when things go very, very wrong.

And then people jump up and down saying "How could this happen?!"

[warriorsavant]

Initially interpreted "Canbera Python Group" as being fans of "Monty."

[tcpip]

Some of them are probably across the lake in Parliament House. Except they don't realise that they're members.

[warriorsavant]

😁

[dorchadas]

That hotel room is positively luxurious. I guess if you have to be stuck somewhere, that's a great place to be stuck.

[tcpip]

Yeah, it was pretty comfortable and very well located. Very good value for money overall. The carpet wasn't perfectly clean, but mine at home isn't either :)

I don't have any complaints about the hotel really, just the fact that Canberra is not exactly the most interesting place in the world to be stuck in, and that the airline didn't have a flight for a couple of days.

[fings]

I liked talk slides. I read a paper on posits last year, and find the concept interesting. I hope they make it into production, assuming they truly hold up to the promise they seem to offer.

[tcpip]

Well, at least posits will be easier to implement in hardware than unum v1, although the latter do have the best error-checking. The "tapered accuracy" of posits seem to be at least something of a plausible compromise.

I have heard on the grapevine that Intel is seriously sniffing around this. Which would be good, because it's been a very bad year for them.