---

ISC Tutorial Day: Sunday

Knowledge. As a relative newcomer to the field of high performance computing (HPC), this is primarily why I attend. The last few years have been a flurry of HPC conferences, workshops, and Linux Cluster Institutes as I try to educate myself and become more conversant about supercomputing. Repetition is helpful, since it is easy to forget from one conference to the next, but so are the different perspectives the venues bring, which is a major reason why I am here for my first ISC.

Tutorial Day, of course, is all about learning. As a newcomer, any tutorial with “Getting started,” “Beginner’s,” “Basic,” or “Introduction” immediately catches my eye. Hence, choosing the morning tutorial “Getting started with Containers on HPC through Singularity” was an easy choice. For me and about 100 other people because that tutorial was packed…a fact that the presenter, Carlos Eduardo Arango, noted with some amazement.

I am interested in implementing containers on our small system. We have reached a point where users are starting to prototype applications on their desktops using a different flavor of Linux, and I’m then stuck trying to figure out how to install it on the cluster. I have been to talks on containers before, but I would have to say that this was the most lucid and compelling explanation I have heard. There appears not to be much of a performance hit, and perhaps a bit of a performance boost, according to data in the first presentation, since users can employ libraries optimized for their applications rather than those installed on the bare metal. On the cluster, users run in their own account, using the batch scheduler. It seems like a perfect solution, and I can’t wait to get home and try it.

Lunch allowed some time to get a bird’s eye view of the ISC Exhibition Floor setup. What will be even more amazing is to watch how fast it all comes down the minute it closes!

For the afternoon tutorial I attended “Better Scientific Software.” I do not write software, at least not at the moment. However, as an X-ray crystallographer, I am a longtime user of scientific software since the crystallography field has been involved with computing almost since the advent of electronic computers. (Crystallographer David Sayre, for example, was on the original team that wrote the first Fortran compiler). Therefore, I’m interested in the process of writing scientific software…especially better scientific software.

The tutorial started with a description of “heroic programming,” in which a single programmer puts in the extra effort to finish a project, and the resulting greater cost in technical debt that increases the future work required to sustain the software. Crystallographic software has suffered the parallel problem of single individuals who have developed and maintained the mainline programs…and nobody else really knows the code. With these individuals now in advanced age, the crystallographic community has made a decided effort to develop new software packages in a more sustainable way.

The central part of the tutorial surprisingly dealt with the soft skills of people and project management. As a participant in the University of Oklahoma-sponsored Advanced Cyberinfrastructure Research and Facilitators Virtual Residency, which also focuses on developing soft-skills, I found this part of the tutorial particularly useful. I was unfamiliar with the Kanban method for tracking workflow, and most critically limiting the number of in-progress tasks. While this was presented in the context of software development, it well applies to other projects, and was my biggest takeaway.

Oh, and I was happy to see crystallography rear its (in this case ugly) head:

 

 

 

 

 

 

Monday @ISC18

Fun. To quote the leaders of the Orientation Session: “A conference without fun is no fun! I would add that the funnest day is the first day. The entire conference still stretches out ahead. The initial keynote speech is almost always motivating. The exhibition begins, and you are in a new place.

Some years ago I was at a meeting in Florence, Italy. A Nobel Laureate was delivering a keynote, and at the very start told us that we should take some time while we were at the conference to experience and explore Florence. Furthermore, that a hallmark of a scientist is curiosity, and you should be interested in exploring the history, culture and good times that a city has to offer. I was so glad to hear the organizers today tell first-time attendees to spend some time enjoying Frankfurt. I certainly plan to do so, as time allows, and will post pictures on the last day.

But speaking of fun, the keynote address by Maria Girone, CTO of CERN openlab, was engrossing. As a domain scientist in chemistry and physics, it was especially exciting to me to hear about the state and future of one of the most exciting science centers on the planet. One can read an excellent summary of the talk by Elizabeth Leake (STEM-Trek) that was featured by TOP500. I will simply share the slide of CERN superlatives: the fastest racetrack, the most powerful magnets, the most sophisticated detectors, the highest vacuum, the hottest spot in the galaxy yet one of the coldest.

The TOP500 list is uniquely paired with ISC, which I learned for the first time at SC17. As I had hoped to attend ISC18, I knew the dates and location when I visited the TOP500 booth at SC17. Interesting things happened when I told them the dates and location of ISC18.  ISC19 will be June 16-20 in Frankfurt, Germany. If you plan to attend SC18 in Dallas, this might be useful information when you visit the TOP500 booth.

The number one supercomputer on the TOP500 was formally announced: Summit at Oak Ridge National Laboratory (ORNL) is number one in both HPL and HPCG, and reasonably green. The press release has more details. As this is the 25^th anniversary of the TOP500, a wealth of historic data were presented, including this plot that neatly shows the legendary “end of Moore’s Law.” The panel concluded with a discussion of what’s wrong with the TOP500, with some very pointed comments.

This discussion continued into the Vendor Showdown, during which vendor presentations were followed by additional probing; pointed questions that really put the vendors on the spot and, in my opinion, gives this meeting an air of collegial debate, which is what I would expect of a scientific conference.

I will keep the vendor confidential, but will give a shout out to Iceland which was represented at the vendor showcase for the first time and has the highest ICT literacy in the world!

Monday-Tuesday, Technical HPC Sessions

I took time on Monday and Tuesday to immerse myself in a series of technical HPC sessions. To steal somebody else’s joke, A few years ago I couldn’t even spell HPC. Having been a Linux user since 1994, the OS/software side of HPC is familiar, but the bewildering array of hardware is not.

It’s getting easier. At my first supercomputing meetings I would have to go around the floor asking vendors, “What it is that you *do* exactly?”  The landscape is coming into focus, but still….

I attended “What’s new with Cloud Computing for HPC,” and “Visualization and HPC” sessions on Monday afternoon, ”Pushing Digital Computing to the Limits,” and “Quantum Computing” on Tuesday morning. These sessions dovetailed nicely with the excellent Tuesday keynote on photonics by Keren Bergman.

I sensed some tension between traditional HPC and cloud computing, which leads me to wonder if our institution should invest in HPC, or simply move to the cloud?  The terms “edge computing” and “Internet of Things” seemed to crop up frequently. My main takeaway is that cloud computing is just another tool which is useful in some situations, while traditional HPC is more appropriate in others; particularly, the “Laws of Physics”-type codes (to replicate a comment from the vendor showdown). The initial freakout over the Cloud, at least in some quarters, seems to have abated and it is gaining acceptance as an add-on to a continuum of computing resources.

I love Eugene Wigner’s famous quote, “It is nice to know that the computer understands the problem, but I would like to understand it, too.” To me, this gets to the heart of visualization, which has long been used in my field, and it’s exciting to see the new, dedicated viz facilities and tools that are now available. Honestly, I was at this session mostly to look at the cool animations and pictures. My favorite was the visualization of flame combustion, shown below, which, as a chemist, intrigues me…and is a visualization I have not yet seen in the community. This is  a followup item.

“Pushing Digital Computing to the Limits,”  was the session I feared the most. But it turned out to be chocked-full of physics, especially thermodynamics–how much energy it takes to compute, or more specifically, move information. This carried over to the excellent Bergman keynote talk on photonics.  Fermions and bosons: Yes!  Solitons?! I about fell out of my chair at that point. Yeah, I understand this stuff!

With the much discussed “end of Moore’s Law” improving data transfer, and reducing the energy associated with data transfer seemed to be common themes for pushing the current limits on computing. The talks in this session advocated various technologies, GPUs, FGPAs (which I sort of understand now!).  But I thought this slide was the most appropriate:
Computational chemistry is said to be the “killer app” for Quantum Computing, as a feature cover article in Chemical and Engineering News declared six months ago. There are issues with tractability as problems get larger. At a recent conference, someone commented that the wavefunction description is inappropriate for large systems. The Slater determinant for a system of 1000 atoms involves more terms than sands in the Sahara or atoms in the universe (I forget the exact comparison). Of course, sixty years ago, the conventional wisdom was that calculations would be impossible for systems with more than 20 electrons. The promised exponential increase in computing power with relatively few qubits is certainly attractive in this context.

I’m fine with the Quantum in Quantum Computing; it’s the Computing part where I get a little lost.  While its being explained to me, I can say that I understand, but afterwards, I am still left with questions. I felt reassured by  Thomas Sterling’s (Indiana University/CREST) admission during the final keynote that even he struggles with understanding QC (he said that QC is what they teach in Computer Science at Hogwarts). 

One aspect of the presentations that struck me is that some qubit devices are fabricated on the mm scale, opposing the movement to ever smaller circuit elements in silicon computing. As I listened, I began to realize that this is a field that cannot be fully understood until you actually do it. And you can now through tools such as QISkit:

The ISC Exhibition

Start your Engines! Or at least your clusters…

The exhibition is a great place for free food, souvenirs and diversions. It is also a great place to learn outside of the regular sessions. I used to feel a little guilty about spending time at the exhibition; it felt like I was cutting class.

Actually, I am here as the result of an exhibition. Seven or eight years ago, back when I was struggling with how I could introduce computation and computational chemistry into our curriculum and into my research, I was going through the exhibition at an American Chemical Society meeting when I happened upon an HPC vendor. My discussion with the salesperson cleared up most of my questions, and I ordered a system from him a couple of years later. A year or two after that, he encouraged me to attend my first regional supercomputing meeting.

I have learned so much at exhibitions that now when I am able attend a conference, I always make sure to block significant time for the exhibition. Besides, after listening to a few talks, I need to engage my brain in a different way.  Hey! Let’s go to the exhibition!

Now, after all of that high-minded talk…I made sure that my “Tour of Chinese Exhibitors” card was completely stamped so I could get my souvenir doll.

To be vendor neutral, I will feature a cross section of the non-commercial booths at the exhibition.

1) The only US university (and as far as I can tell the only university from the Americas) to make an appearance at the exhibition was the Louisiana State University Tigers!  They have exhibited at ISC before, but not last year.  Fellow US residents: Thanks for representing!

2) PRACE is the European equivalent of XSEDE in the US, and includes many of the European institutions at the exhibition.  They enjoy warm relations with XSEDE and XSEDE Forum Chair John Towns.

3) The Irish Center for High End Computing operated an extremely friendly booth. They have HPC systems in Dublin and Galway, and will be bringing in a new $5M system this summer. They report their heaviest workflows to be weather and oil and gas. Computational chemistry also had good representation at the booth!

4) Baden Württemburg HPC operates a state-wide network that reminds me of similar networks in the US, such as the venerable OneOCII network in Oklahoma or the ShowMe CI network in my home state. In this case, however, machines at different institutions are dedicated for use in different domain areas, unlike the state systems in the US which are for general use.

5) EPCC in Edinburgh collaborates with PRACE to provide HPC training and operates an impressive pre-college outreach program

6)  The Korean Institute of Science and Technology Information (KISTI) just brought it on at their booth.  That’s all I have to say.  If you missed them, you really missed out. Oh, and Korea; congrats on your World Cup victory that week!

Cluster competitors were still cooking….

7) The National Supercomputing Centre in Singapore has four major stakeholders, including three universities and a research institute. It runs a varied workload. They are active in regional cooperation with fast data links to the US and Japan.

Editor’s note: Singapore Director of Industry Innovation Laurence Liew’s presentation on the “Smart Cities” panel discussion was interesting; they are investing $250,000 apiece (+ matching industry funds) to retrain up to 200 engineers to specialize in AI.

8) The Center for High Performance Computing (CHPC) in South Africa is a group that I know, personally, from having attended STEM-Trek workshops! Their Lengau system is number 246 on the TOP500 list, and they are the broader sub-Saharan regional leaders in the development and promotion of HPC.  Their Student Cluster Competition teams also have a strong record at ISC.

9) Our last stop (and continent): The Pawsey Supercomputing Centre in western Australia. They have two systems on the TOP500 (a cloud provider is #200 & Magnus @ #217); they operate the only Cray systems in Australia. Given Australia’s large mining industry, workflows lean toward supporting geology and mining. They actively cool their systems with ground water: pumped in at 21° C and re-injected at 23°C.  A July 5, 2018 HPCwire feature by STEM-Trek’s Elizabeth Leake describes the Australian cyberinfrastructure in detail.

10) The talk about the Summit supercomputer by Jack Wells of ORNL may have been the hottest ticket on the exhibition floor.  I have never seen an audience seat so quickly.

11)  The one thing that could bring the exhibition to a halt that week:

Fade to black (until next year!):

Science, History and more…

Building upon what I felt was a strong science undercurrent to the ISC High Performance conference, I attended three HPC applications sessions on the third day that provided not only outstanding HPC, but excellent science.

The most exciting, given the recent discovery of gravitational waves, was the first titled, “Astrophysics & HPC.” The first talk on “Binary Neutron Stars: Einstein’s Richest Laboratory” provided jaw-dropping animations of simulated neutron star collisions. Neutron star collisions produce a distinctly different gravitational wave profile, as compared to hole collisions. The initial product of a neutron star collision is an unstable High Mass Neutron Star, which then decays to a black hole while ejecting about one percent of its mass into space as heavy elements. Most of the heavy elements, I learned, are formed in this process with calculations showing one neutron star collision yielding six earth masses of gold.

The textbook chemistry explanation that heavy elements arise from supernova explosions needs to be updated! And I always like to bring something back from a conference that I can share with my General Chemistry students.

There’s more. They also took a time-dependent gravitational wave signature and Fourier transformed it into the frequency domain to generate a non-electromagnetic, gravitational wave emission spectrum that warmed the cockles of my spectroscopic chemistry heart:

“Quantum Computing Applications” followed, and I was just along for the ride. This session attracted, in my judgement, the most intense interest with more attendees pulling out their phones to take pictures of the slides than any other session I attended. This would be in line with Thomas Sterling’s comment at the closing keynote, “Quantum Computing: No longer a punchline.” The focus of the session was on quantum annealing, in which the quantum computer is prepared in an initial state and allowed to evolve over time toward a global minimum. The first talks focused on applications in computational chemistry. Again, I am having difficulty wrapping my head around this, but plots like this intrigue me:

There were further non-chemical examples, such as optimizing airport transfers, and election forecasting, to demonstrate a broad range of possible applications.

Last of all, I attended “Materials Sciences and HPC,” which is most closely related to my field. Well, actually, I was late because I got distracted by the Summit Supercomputer talk in the exhibition and came in as the first talk about perovskites was ending, and I was sad because I live for perovskites. The remaining presentations were quite good, with the last two focused on data analytics in materials discovery and simulations to mimic biomaterials. But not only was the conference reaching its end, I was reaching my 28^th day of international travel. So, this last image of the simulation of a probe tip stretching a graphene layer to its breaking point was weirdly appropriate:

This was the fifteenth year Thomas Sterling has delivered the concluding ISC keynote. Sterling reviewed the events in HPC over the past year, and the preceding fourteen years, and gave emotional tributes to leading figures in HPC who had recently passed away.

I felt this was a particularly important part of the presentation. It is easy to forget the names and personalities who have shaped our fields. I was recently at a scientific school where a question was raised about who, exactly, belonged to two initials and a surname in one of the references. The last speaker modified his slides to spend some time to pay tribute to his colleague who had made tremendous contributions to the field and had guided the careers of many current scientists.

Awareness of the history of a field is, I think, an important overall part of one’s knowledge. With 25 years of the TOP500 list and 15 years covered by Thomas Sterling’s keynote bookending the ISC High Performance conference, it uniquely delivered a sense of history along with cutting edge HPC trends and topics.

Summing it all up:

Is ISC what I expected? The answer is no. Looking over the program before the conference, I had expected something like the SC conferences I have attended. There is a great deal of similarity in structure, don’t get me wrong, but ISC has a very different feel.

It is smaller than SC, although growing with a record attendance this year of 3,505, a record-setting exhibition, and a selective acceptance rate for papers. Yet it has a distinctly different feel and approach. There is a more informal, more intimate feel to the keynotes and the sessions and, unsurprisingly, it’s far more international. My initial assessment that ISC has the flavor of a scientific conference was fully realized as the conference progressed.

At the same time, it did not feel like a small conference. I only experienced a cross-section of it; I skipped Machine Learning Day, Industrial Day, etc. and focused on areas that would benefit me the most. Yet, I had no difficulty filling my “dance card.”

Is it worth attending both ISC and SC? Definitely. Both are great conferences, but they are not duplicates of each other. For a US-based attendee, ISC is more of a financial challenge, but not overwhelmingly so. I piggybacked my visit to Frankfurt with another conference I was attending in Europe which reduced the overall cost.

But with careful planning, it may not cost much more for U.S. scholars to attend ISC. Frankfurt, where ISC is held, is home to a major European airport that offers many direct flights, and is served by low-cost airlines. I flew one such carrier home, and while there were some inconveniences and headaches, my luggage and I arrived together, and on-time. Conference lodging is comparable to that at SC.  But, there is no ACM discount for registration, which may be a hardship for some.

The current conference venue is conveniently located and quite pleasant. There is a multistory mall directly across the street with a wide variety of restaurants, and a “drug store” that sells sundry items right inside the entrance.  Actual drugs, including pain relievers which are kept behind the counter, are sold at an apothecary shop…which is also in the mall. Most critically, there is a luggage store on the second floor in case you pick up too much SWAG in the exhibition, or overbuy German souvenirs.

A short walk in the opposite direction leads to a park and green space. Beyond that is a quiet residential neighborhood, where you will find an Italian restaurant with outside dining every few blocks and a small grocery market. Walking through this neighborhood leads to Goethestrasse, which is lined with high-end shopping boutiques, and leads to a broad plaza with a statue of Goethe himself (born in Frankfurt!).

Walking even further and bearing to the right leads to the Römer district with its painstakingly reconstructed “old town” buildings (most were destroyed during WWII) housing a variety of souvenir shops and sidewalk dining.

The Römer is just a short walk from the River Main, which is crossed by the pedestrians-only Eiserner Bridge which is adorned by padlocks that were placed on its frame by lovers.

The opposite bank contains a nice beer garden and restaurants on barges that make for a pleasant evening visit.

The bridge itself is bears a sign in Greek with a quote from Homer: “While sailing o’er the wine-dark sea to men of strange speech.”

This could be a nod to Frankfurt’s position as a center for international banking, with 40 percent of the city’s population being non-German. In my evening walks around the city, I overheard many English conversations, and I did not experience a language barrier at the hotel, or any of the restaurants or stores that I visited.

Mark your calendars!

ISC High Performance 2019 is June 16-20, 2019 in Frankfurt! You won’t miss this exciting opportunity to find out what colleagues from around the world are doing to promote performance technologies.

Feature photo of ISC18 laptop bag by Bond:

As a side note about the SWAG, Bond said,

The conference bag is one of the nicest I have seen. It is attractively designed, relatively sturdy, and large enough to be useful—but not so big as to be cumbersome. It will get a lot of use this fall taking papers home to grade.”