Looking back at the training week

An alley behind the Cathedral of the Holy Cross and Saint Eulalia

During the last few weeks before the Summer of HPC, I literally didn’t stop for a second. I was exhausted after trying to finish my exams so I was really looking forward to some time off. And what can be a better place for a vacation than the sunny beaches of Catalunya!

I didn’t have many expectations about the training week, I was really looking forward to the lectures and practicals, maybe to do a bit of sightseeing in Barcelona and most importantly to pay back my sleep debt. In the end, the lectures and practicals came out even better than expected and I did more than a bit of sightseeing, but I couldn’t have been more wrong about getting some rest!

The training week started on Monday 29th of June, we met with the head of Summer of HPC, Leon Kos, and other participants in the morning and, fortunately, didn’t start with the training straight away. Leon took us for a nice walk around gardens of Palau Reial de Pedralbes. Even as soon as 9 am the air was hot in the city and the shades of the trees came in handy. We also used the time to get to know each other. Or at least get to know each other’s names. Or at least the countries we are from and where we are heading to…

We were all going to spend together just the training week, only 5 complete days, and then depart to different institutes for the rest of the summer. Nobody was going to spend their summer alone, though, each institute took in two or three of us. Obviously, this was the person everyone was looking for at first. It’s not an easy task when there are twenty complete strangers. I remember asking everyone I was talking to, “Do you know who Paolo is?”, but nobody seemed to know him. I succeeded in the end and was not disappointed by the guy I was going to spend the summer with. It must be said, I was not disappointed with anyone, just on the contrary, everyone was much more fun and much friendlier than I could have imagined.

After enjoying the peace of the gardens, we had a late breakfast and met the rest of Summer of HPC organizers in Barcelona Supercomputing Centre (BSC). There was a short welcome session, amazing lunch and talk about PRACE and the Summer of HPC programme. It would be wrong not to mention here, that PRACE is paying for the whole Summer of HPC and without this support, many of us could not afford it.

In the group, there were people from thirteen different countries, but apparently the Irish guys had the lead for the first night as from all the pubs in Barcelona, we ended up in a proper Irish one. We had a beer or a cider, watched some rugby on the tv and debated about the world’s problems as people usually do when they know each other for just a day. Or that was maybe only what happened at our table…

In the end, I was happy to go to bed knowing that I will probably not be bored with these people, even if it was only for one week.

I will not write about every day of the training week as much as I’ve written about the first one, but there is certainly one thing that must be mentioned that happened during the second day.

Mare Nostrum, credit BSC, for more pictures visit http://www.bsc.es/about-bsc/gallery/interior-torre-girona

Again, we did not start with the lectures immediately after breakfast, but we visited the so called Mare Nostrum, the largest and most powerful supercomputer in Spain. It’s not the most powerful supercomputer in the world (world rank 77 as of July 2015) but it is maybe the most beautiful one. It is located in Torre Girona a mid 19th-century neo-romantic chapel. It is a very peculiar place; the futuristic machine sealed in an air-proof glass container with flashing diodes and colorful cables in contrast to the dimly-lite hall with pale stone walls and ornate windows with a chorus music playing low in the background. There are many beautiful churches in Barcelona, but I was not as impressed with any of them as I was here.

In addition to the charming main hall, there is also a room dedicated to the history of supercomputing in BSC. We saw a car sized supercomputer that was state of the art about twenty years ago while having approximately the same computation power as tablets have today. This gives you a much better notion of how large a step technology has taken since then, than just comparing some numbers.

Mare Nostrum

During the rest of the morning and the whole afternoon, we had several lectures about parallel architectures and algorithms and the MPI programming model. The other lectures introduced us to some more advanced tools being developed in BCS such as Paraver, Dimemas, and Tareador. As it is common in a scientific community, the names are not given by chance. Paraver is a tool for visualization of parallelism and can be used to discover performance issues and “para ver” means “to see” in Spanish. Dimemas is a performance analysis tool for message-passing programs which generates data to be used by Paraver and “dime mas” can be translated as “tell me more“.

It was very interesting to use these analysis tools. One can write a parallel application to employ e.g. all four processors of your laptop. But it is no use if only one of the processors is active at a time and the other three are just waiting, doing nothing. And that is exactly what Paraver and Dimemas can discover and help us deal with it.

Tuesday evening was not lead by the Irish guys, this time we decided to do a bit of sport-related activity after sitting all day behind our laptops. We were told that it is a nice, twenty-minute walk from the hotel to the top of one of the hills above Barcelona. The voting was unanimous so we departed with a vision of us looking down at the never sleeping city. One and a half hour later, having cursed the guy who thought of this, exhausted and dehydrated, we crawled to the top but soon forgot about everything else. The view was spectacular, the night was warm and we had beer with us, of course…

On top of the hill. Credit: A random cyclist on Sarah Jenkins’ phone

On Wednesday, we had more lectures and practicals about MPI and OpenMP programming models. They are the two most common standards used for parallelisation. MPI stands for Message Passing Interface and OpenMP for Open Multiprocessing. The biggest difference is that MPI was originally designed for distributed memory systems (such as clusters) and OpenMP for shared memory systems (such as your laptop). There are numerous factors influencing which of the models is better in a specific situation and they can also be used simultaneously. I’m not going to write about these here more, but I think that for a computer scientist or any other technically educated person, these are things worth reading a bit about.

The gang. Credit: BSC

After the lectures, we got our Summer of HPC t-shirts, took some pictures in front of the Torre Girona chapel and headed to the beach. There we participated in the shooting of a short promotional video for the Summer of HPC. We were a bit afraid at the beginning, but in the end we enjoyed it. Especially our attempt to make a human tower, as it is customary in Catalunya. I admit it wasn’t as high as the towers you can see at festivals, but It was fun all the same. Check it for yourself, it’s just at the end of the video.

After the shooting, we tried the water (it was good) and stayed at the beach and chatted also with the guys from BSC who lead the lectures and practicals. It was nice to get to know them and hear what are they working on when not explaining the basics of high-performance computing to us.

On Thursday, we were introduced to the OmpSs which is a connection of OpenMP and StarSs, another tool developed by BSC. The next thing on the menu was so called hybrid programming (usage of MPI and OpenMP at the same time) and CUDA. All of these programming models represent a field so wide, that you can study it for years and still not know everything. Therefore, all the talks were just an introduction and the practicals were only supposed to give us a feeling of how it works. Nonetheless, it will be much easier to try and learn this for ourselves when we actually need it.

In the evening, we went for a walk through Barcelona old town. We took a subway to Plaça de Catalunya, walked through the famous La Rambla, which is by the way very overrated, to the Port of Barcelona. On the way, we passed the Columbus Monument, which is a great pillar with a statue of Christopher Columbus on top and surrounded by huge statues of lions. We couldn’t resist taking some pictures there too. From the port, we wandered through the curvy narrow streets of the Gothic Quarter, saw the old Barcelona Cathedral (the proper name is the Cathedral of the Holy Cross and Saint Eulalia) and then continued in the direction of the Arc de Triomf.

The Columbus Monument. Credit: Sarah Jenkins

Our efforts to get there were, however, sabotaged by a nice, finally not Irish but Spanish, pub we met on the way. We all had a beer or a glass of wine, which was very refreshing. Every day I was amazed by how hot it actually is in Barcelona even during night. After restoring our strength, we started to head home and since it was after midnight, and the subway does not work that late in Barcelona, we decided to continue just by foot. It was supposed to be just about five kilometers and we had a lot to talk about. The five kilometers turned out to be quite a task after the whole long day and we had to stop once more, this time in an English pub. In the end, however, we got to the hotel safe and sound with just enough time to be able to get up for the lectures one last time.

The lectures on Friday were different from all the other days. It was all about visualization of scientific data – a subject most of us are dealing with during our work on projects for the rest of the summer. The lectures consisted of an introduction to some of the most widely used software – VisIt and ParaView. We are using some results produced by OpenFoam, which is an open source computational fluid dynamics (CFD) package. As an example, we used a model of a motorbike with a rider, used the OpenFoam package to compute how the rider and the bike respond to wind and visualized the results in ParaView. All in all, we were able to simulate a wind tunnel on our laptops and see the results within one hour, which was quite impressive. Of course, this was just a tutorial and the actual work with all of the software packages is much more complicated.

Motorbike visualisation in ParaView, OpenFoam. Creadit: vespalabs.org

For the last evening, we went for a dinner to a nice restaurant near our hotel. The food was great and it was a proper way to end the whole week. During the five days, I made many new friends and I have twelve countries I want to visit and meet again with the great guys of the Summer of HPC.

I would like to thank Leon Kos, and all guys from BCS and PRACE for the amazing week they prepared for us.