Archive for the ‘conferences’ Category

Post SC07

November 17, 2007

Supercomputing 2007 was held in Reno, Nevada.

  1. Day One: November 10, 2007
  2. Day Two
  3. Day Three
  4. Day Four
  5. Day Five
  6. Day Six

Here are some scattered thoughts on my SC07 experience. . .

  • The student volunteer (SV) program could be more organized. Compared to my SV experience at SIGCHI07, the SC07 job assignment process seemed arbitrary. At SIGCHI, SVs use a web interface to rank their preferences for the next day’s available volunteer positions. Every night, the preferences are compiled, and job assignments are made accordingly. I think SIGCHI’s laissez-faire assignment system yields high SV satisfaction. On the other hand, Supercomputing-07 SV jobs were assigned by the SV organizers. The organizers are great people, and I believe they made a fair attempt to judiciously assign jobs. That said, there was a lot of job-swapping among the volunteers. I realize that SIGCHI and SC are radically different conferences. However, both conferences share a similar scale of organization and I think the Supercomputing SV program could learn a lot from the SIGCHI SV program.
  • Although attending a conference with my cohort is nice, it also breeds xenophobia. At evening events, I found myself tempted to just hang-out with the U.O. crowd.
  • SC07 feels more like a trade show than an academic conference. The exhibit floor is HUGE, and most of the booths belong to companies hawking high-performance computing hardware. Although commercial activity is necessary for SC to be shiny and big, it thwarted my ability to meet other students, faculty, and researchers. Specifically, most of the evening parties were dominated by sales reps, not researchers.
  • I highly recommend lodging in a hotel across the street from the convention center. In the mornings, I had no trouble waking-up and attending keynote lectures, whereas my colleagues faced a thrity-minute shuttle ride across town.
  • Final thought: slot machines are like Hell for a light-induced epileptic.

SC07 Day 6

November 15, 2007

This morning, David Shaw gave a talk titled “Toward Millisecond-scale Molecular Dynamics Simulations of Proteins” Last year, David’s group released Desmond, a computational simulation of molecular dynamics. Desmond has many applications, including simulations of protein fold behavior. This year, David’s talk focused on the future of Desmond, which apparently involves Anton, a supercomputer specialized for running Desmond.

After a long lunch with Maryrose and afternoon chit-chat with the ASC folks, I found myself at the Grand Sierra Resort for the SC07 Technical Party. Blue Man Group (BMG) performed a short 30-minute show, which was more entertaining than I expected. In a nutshell, BMG is like the offspring of french performance art with American rock and roll.

SC07 Day 5

November 15, 2007


SC07 Day 4

November 13, 2007

Neil Gershenfeld, the head of MIT’s Center for Bits and Atoms, gave today’s keynote address. To summarize Gershenfeld’s lecture: the killer app of digital fabrication is personal fabrication. Gershenfeld highlighted the MIT FabLabs, and gave examples of boundary-breaking personal computation: wallpaper computers, analog computers, and ad-hoc clusters. This is one of the most compelling keynote lectures I’ve seen. If you’re interested in Neil Gershenfeld, click here to watch his 2007 TED talk.

Alexandros P. Stamatakis presented a paper titled, “Large-scale Maximum Likelihood-based Phylogenetic Analysis on the IBM BlueGene/L”. Stamatakis and his team created a parallel implementation of RAxML on the BlueGene/L. Although I’ve used several software packages for phylogenetic tree construction, I was unaware of RAxML. According to Stamatakis’ publications, RAxML is qualitatively comparable and computationally faster than my current software of choice: Mr. Bayes and PHYML. After hearing Stamatakis’ presentation, I’m interested to use RAxML in one of my current projects, which requires the construction of thousands of phylogenies.

In a related paper session, I learned about the BlueBrain Project, an attempt to computationally simulate every neuron in a mammalian brain. The BlueBrain Project is very ambitious, given the computational complexity of mammalian neurology. (See the image below).

Later. . . . SGI hosted a party at the National Automobile Museum, and SiCortex hosted a party at the National Bowling Stadium. Basically, “cars and bowling” sums up Reno.

SC07 Day 3

November 12, 2007

At the Workshop on Grid Computing Portals and Science Gateways, Marlon Pierce talked about “Web 2.0 for e-Science.” Marlon’s talk was a shotgun blast of compelling information. He talked about micro-programming versus macro-programming, mashups (read more at The Programmable Web), web services, and how computational science can harness the Web 2.0. The big point was that scientific web applications (for instance, GenBank) are the perfect building blocks for scientific mashups. Marlon asserts that most scientific workflows can be implemented with a web mashup, composed of smaller web gadgets and web services. Other errata from Marlon included The Gartner 2006 Hype Cycle. (See the image below).

Later in the day, the exhibition gala was crowded fun. . .

SC07 Day 2

November 11, 2007

The cluster challenge is one of the coolest events at SC07. Six teams of undergraduate students build small clusters on the exhibition floor and then race to complete a long set of computational benchmarks. The teams are limited to 26 amps of electricity, which puts a premium on power-conscious hardware. The benchmarks include standards (like HPCC) alongside GAMESS, POV-RAY, and POP. The organizer of the challenge told me that he doesn’t expect any of the teams to successfully compile and execute all the benchmarks.

Today I helped setup the cluster challenge and chatted with several teams about their hardware. Most teams are using multi-core Xeon architectures, with various flavors of Infiniband. Indiana University is using an experimental Myrinet interconnection, which is probably very fast but a headache for driver-support. Purdue is the only team using an Opteron chipset, which might introduce unexpected electrical limitations. Stonybrook University is using a large number of relatively slow Xeon CPUs (1.86 Ghz), with a total of 100 processing cores. Stonybrook’s strategy might give them an advantage with respect to parallelism.

The annual exhibitor party was held at the National Bowling Stadium, a.k.a. the “Taj Mahal of Tenpins.” This place is so ripe for parody, I’ll bite my tongue. After the bowling fest, I met T.M. from the University of Tennessee. This is interesting: TM is a philosopher who now works in high-performance computing. He recommends the writings of Charles Sanders Pierce.

SC07 Day 1

November 10, 2007

Today I traveled to Supercomputing 2007 (SC07) in Reno, Nevada. (Julian, Heather, Kynthia: Thanks for being great Portland hosts.)

On the flight from Portland to Reno, I met Gayle Gregory.  Thanks for the energy work!

Being in Nevada, gambling devices are everywhere. I am staying in the Atlantis Casino and Resort, which is the mother-hive for shiny gambling machines. The lobby says, “Blink. Beep. Ring. Ca-ching.” Basically, this is a seizure-inducing Hell. (See the photo below).

Tonight was the student voluneer (SV) dinner – which involved an hour bus ride to Lake Tahoe and a dinner cruise on the Tahoe Queen. The dinner was delicious (however, it was a vegan nightmare because none of the food was labeled).  On the return bus ride, I had a good conversation with Sarat Sreepathi about computationally modeling the dispersion of environmental pollutants in bodies of water.  (To be accurate, Sarat deals with the inverse problem of finding the source of existing contaminants). Sarat’s problem is impossible to directly solve, and thus requires sophisticated heuristics which are very similar to problems I face with evolutionary simulations.

Spectral Network Analysis

August 29, 2007

This video is exciting:

Pavzner and his team address the challenge of determining a protein’s sequence, given mass spectrometry data. This problem is challenging because every protein undergoes translational modifications, and therefore mass-specs actually measure proteins which deviate (slightly) from their non-translated nucleotide sequences.

An unsolved challenge is to correctly infer a protein’s sequence, given mass-spec data. However, we CAN compare mass-specs against a database of existing (and known) mass-specs in order to guess the sequences. This process is computationally expensive because it requires a linear search of the database for every queried mass-spec search key.

Pavzner present a more efficient algorithm for “guessing” a protein’s sequence, given its mass-spectrum. His technique involves constructing a network of spectral data, and then using that network as the basis for a search. This is remarkably faster than traditional database searches. Pavzner, et al, apply this technique to whole-genome spectrometric data, and yield promising results.

My obscure notes:

17:00 – Why is the signal-to-noise ratio reduced “six-fold” ?

18:01 – Look-up reference for “anti-symmetric pass approach” to solving an alignment between two sequences of unequal length. Can we use this storage technique for other information domains: phylogenetic trees? Electroencephalographic data?

28:07 – The use of “snake venom” makes any science project sound cool.