Archiv für den Monat: September 2013

Week #100

(Date: 23.9.2013 – 27.9.2013)
Wohow! Week 100. 100! Congratulations to me. How time passes. Maybe I will look back at the last 100 weeks in a dedicated posting. For now just: Wohow!

This week I did three things:

  • Experiment Beam Time. An experiment of our institute, a kind of pre-experiment for PANDA, has beam time at the local Jülich accelerator COSY. They need help and so I’m sitting there looking on screens for nothing to go wrong. Not much to do for me besides that, actually. Looking at the data acquisition and restarting it if it crashes. That’s it.
  • Preparing a Talk Proposal. I want to present my GPU research at a conference next year. For this, I need to submit my talk proposal. Abstract and description have to be written, summaries and documents assembled. Wish me luck for my proposal to succeed!
  • Prepare the Day of the Open Door. Sunday (29.9.) is a day of the open door at our research center. I volunteered to show some things we are doing at the institute. Tracking, of course. Together with a colleague I prepare some slides to explain the concept to non-scientists coming by on Sunday.

Talk: Status of GPU Tracking Algorithms

Slides: 31
Duration: 45 Minutes
Occasion: Report at Weekly Meeting
Location: Home Institute, IKP-1, Forschungszentrum Jülich
Date: 19.09.2013
Background of Audience: Hadron physicists

Basically the same talk I held at the PANDA meeting in the week prior to this week. Just some additions and simplifications (not all people in the audience are PANDAs — a lot of them, though), as well as corrections and improvements.

In the first half there were a lot of questions and discussions on specific things of my algorithms. So my time ran away fast. I decided to skip a few slides in the end to finish a bit faster. But that’s ok for this meeting.
After the meeting I discussed some stuff with two colleagues. Productive!

View on Scribd, local PDF.

Week #99

(Date: 16.9.2013 – 20.9.2013)
The week after the PANDA meeting.
As usual, this means the following: Filling out paper work for travel reimbursement; Following up on stuff of the meeting.

Also, this Thursday I have a talk in the institute’s weekly meeting. See the post for more info on that.

Week #98

(Date: 9.9.2013 – 13.9.2013)
PANDA week. Meeting in Bochum.
Monday: Arriving and test-talking our talks. Quite a strange thing we did there. We just go quickly through everyones slides. There is next-to-no feedback to mine. Hmm.

On Tuesday I wanted to listen to the talks in the DAQ session. But unfortunately, the university lacks proper WiFi in their lecture rooms (as well as proper power supply). And since I have to finish up my talks, I am forced to leave the session and work in a WiFi-enabled seminar room. Don’t like that.
I create two ROOT macros (1|2) which break down PandaRoots simulation chain to the very basic necessary to have a complete, stand-alone example running for my CUDA+PandaRoot tutorial.

Wednesday is the horror day: First, some plenary talks. After that the tracking algorithms session which I start off with my overtimed talk. Then: Everyone else also takes too long. Instead of having a lunch break from 13:00 till 14:00 we end up having a fast food lunch break from 13:45 to 14:15. The next session starts at that point and, again, everything runs out of time and some talks even take 50% longer than they were scheduled for (not mine, though). A discussion of tracking algorithms was planned starting from 16:30. It doesn’t start until 18:00 because of the talks from before. At 18:50 we have to leave the lecture hall — and the discussion ends up being very short. The condensed result I put into our wiki.

Fortunately, at 19:00 the social activities start and we have a very nice and refreshing evening with the whole collaboration.

Thursday is also a bit busy but finally is alright. Our Young PANDA discussion is too short for me, but my train leaves already too early. Sadface.

Talk: Using CUDA from PandaRoot

Slides: 9
Duration: 7 Minutes (from originally 10 min)
Occasion: PANDA Collaboration Meeting 3-2013, Computing Session (both links are password protected)
Location: Ruhr-Universität Bochum
Date: 13.09.2013
Background of Audience: PANDAs, having something to do with computing.

Short update on the slides I presented during an online meeting a while ago. This time: Updated to new, more automated CMake structure.
Since the session began later and we were already running into overtime, I just went through the slides very quickly.

View on Scribd, local PDF.

Talk: GPU Tracking Activities at Jülich

Slides: 30
Duration: 35 Minutes (incl. 10 min overtime)
Occasion: PANDA Collaboration Meeting 3-2013, Tracking Software Session (both links are password protected)
Location: Ruhr-Universität Bochum
Date: 13.09.2013
Background of Audience: PANDAs. Fellow tracking algorithms developers. But also some other guys sitting in, since there was no real session parallel to this one.

The talk itself was good. I wanted to have a talk with everything in to summarize what we’re working on.
Unfortunately I didn’t test talk it, so I didn’t know how fast each part was down to the minute. So, in the end, it took me quite some more time then originally planned.
A few factors were responsible: Probably also for the audience, everything I presented was quite interesting — no one interrupted me when my time was over1. The session was very casually organized, but in the end quite fully-packed with talks. And the discussions directly afterwards were interesting but also a bit off-topic. And long.

I was not the only one taking over time. Every other talker during the session took their 5-10 minutes extra. A strange session it was.
Anyway: I hate people taking too long for their talks (especially that much). It won’t happen again for me. I promise!

View on Scribd, local PDF.

  1. When session conveners are too involved… 

Week #97

(Date: 2.9.2013 – 6.9.2013)
The main topic of this week is the preparation of my two talks for next week’s PANDA collaboration meeting in Bochum.

To get a bit into the status of the GPU port of the Triplet Finder algorithm by the NVIDIA Application Lab, I meet with the guy from there on Monday. He shows me around in the code and together we brainstorm, how to continue the work and what the next steps are.

On Thursday I do something crazy: I drive to work by bike. Fun, but a bit exhausting (2 ⋅ 35 km!). Convenient that this is the day with all the meetings. First off, PhD physics seminar, after that IKP and PANDA meeting and then Andrés test talk.

On Friday we have some summer festivities of the institute. It’s nice but, one could have guessed, I’m not yet ready with my talks for the next week. So I can’t celebrate but have to continue preparing slides. Well. Fine, I guess.

For references in my talks I upload all the current GPU tracking algorithms to the PandaRoot SVN. The location is

During the week I’m in correspondence with a software developer of FairRoot. I want to update my CMake+CUDA+PandaRoot example to our new CMake structure but fail in doing so my self. On Friday, just in time, the developer finishes. The new structure is much easier — although because of compiler incompatibilities between nvcc and clang, at least on OS X, there is some overheading steering code necessary.
Anyway: It runs.

Week #96

(Date: 26.8.2013 – 30.8.2013)
Really nothing worth talking about.

The week is full of »Hey, I’ve got a great idea of how to find multipeaks in this histogram« continuously followed by »Aaand it doesn’t work«s.

We have another idea of generally changing the Hough Transform approach — like a Hough Transform of the second degree. But when doing this, one needs to use the measurement uncertainties of every hit point. And this complicates everything so much, again, that also this idea is very time-consuming to look into. Meh.
So, yeah. That’s it.

Oh, and I spend some time on documenting how to use the LaTeX plugin for Sublime Text, LaTeXTools, with the keyboard.