(Date: 12.5.2014 – 16.5.2014)
Starting early on Monday morning, I’m traveling to Frankfurt. From Monday to Wednesday, there’s a workshop at the Frankfurt Institut for Advanced Studies, FIAS, on particle tracking & Co. Together with my new colleague working on GPU-based tracking and a few other PANDA members we represent our experiment. I hold a talk on our GPU stuff, there’s a PANDA introductory talk and a PANDA talk on FPGA-based tracking.
People seem to be interesting what we are doing at PANDA. They like that there are so many contributions from us, being at this workshop for the first time… Yay us.
Presented are being the statuses and challenges other particle physics (HEP & hadron) experiments are facing. Also, they talk (a lot) about software packages, especially doing some kind of vectorization. Very deep-level and I spaced out quite soon – but at least I have a picture of the base-level parallelization efforts people are doing.
All in all the workshop was quite nice. Very expert-y. And an interesting discussion at the end.
On Wednesday, at 13:30, after the workshop, I don’t leave heading home but to Dresden. There’s another workshop on Thursday and I got invited to talk. After some self-induced train mixups I am finally at TU Dresden’s guesthouse at 21:00 in the evening, giving the final touch to my presentation with tired eyes.
The workshop in Dresden is hosted by the CUDA Center of Excellence, a kind of virtual institute to collect all CUDA efforts by some working groups using CUDA in and around Dresden. The title of the workshop is »Fast Data Processing on GPUs«, meaning everything and nothing – but that’s ok.
There are talks on biology, software architecture, image processing – and mine on physics. I present pretty much the same talk I showed at GPU Tech. Nothing new since then and the same focus group.
Only difference: The organizers are holding a session called »Bring your own GPU challenge«.1 I am one of the only ones prepared and present my problem on parallel multi peak finding in the Hough space.
After explaining my problem and de-confusing everything said to short2 there are some good remarks from the expert audience. I note them here to have a more deep look into at a later stage when I have again time for something like that:
- Intersection of Sinoids: For every \((x,y)_i\) point pair, create a \(r_i(\alpha)\) equation and calculate intersection with a different \((x,y)_j\). Find peaks then (should be clusters). I had this idea before, but maybe with the new techniques I worked on till now, this can work better?
- Adaptive grid, last step: neighbor finding
- Research algorithms able to find »brightest pixel« – because that’s more or less our problem
- Change parameter space to more dimensions and use clustering algorithms there. Something in 3D, probably. Have to think about that and am not sure in detail what the guy offering the idea really meant.
- Maybe of interest: Barnes-Hut algorithm
- Read about Octrees
Safe to say: The Bring Your Own Challenge discussion was the best thing for me in the workshop.
The talks were very topic-specific, I didn’t really understand that much about.
Unfortunately, the last two talks were canceled. Originally, I didn’t want to leave the workshop early to catch an earlier flight home, because being one of the talkers and rudeness and stuff, I booked the next flight home. At 20:45. So now, with the talks being canceled, I had to spend five hours in Dresden. I found a quiet place to sit in the university and did stuff.
Of course, come 20:45 at the airport, my plane has a delay of 45 minutes. Workshop end: 15:00. In my flat: 0:30. Sadface.
On Friday, there’s lots of stuff to do, from which I only remember filling out a number of travel reimbursement forms. It’s the unpleasant that sticks, right?
All in all, a very interesting week. Lots of traveling3, lots of interesting and specialized topics and discussions.