We took a field trip
Shaking things up by leaving our small computers and going to one really big computer
Somewhere between Schrodinger’s Cat and like everything from Interstellar, nestled in the Annals of Misquoted Science Parables lies the tale of Laplace’s Demon.
In 1814, a really, really smart man named Pierre-Simon Laplace, mused: if some agent, or “demon”, knew the position and momentum of every particle in the universe, then by following the deterministic laws of physics this demon could predict their values at any time in the future.
Simple, right? Everything we interact with is made up of tiny particles. They each follow a set of physical laws. So, maybe we could predict the future by simulating the trajectories of every single one of the countless particles that form our environment!
Laplace founded the basis for casual determinism, which is the notion that aspects of our world can be predicted with enough knowledge of the system’s present state and a well-defined set of instructions for stepping this state forward through time.
And so, 200 years later, everyone from meteorologists to cancer researchers have sought Laplace’s crystal ball in some form or another. Could we predict the weather by knowing about every single water droplet in the atmosphere? What about a cure for cancer by simulating a tumor’s behavior under millions of different conditions?
This is the power of simulation, and it’s the reason The University of Texas is spending nearly $60 million on what could be mistaken for a warehouse of refrigerators.
As the thirteen23 team stepped into the room that houses the Stampede supercomputer, we were greeted with a whooosh! of freezing air and the scream of thousands of processors and fans. Decked out in summer clothes, we were considerably underdressed for the occasion, as cozy employees in thick sweatshirts and earplugs bustled back and forth. “When this thing’s running at full capacity, this place starts to get up to room temperature,” Joon, our guide, assured.
Although TACC houses a number of supercomputers, Stampede definitely packs the most punch. It boasts an impressive 6400 nodes, which are essentially desktop computers packaged tightly into the giant arrays you saw above, sandwiched between towers of cooling fans. Across these nodes, Stampede has over half a million processing cores. (For comparison, a high-end desktop processor will typically have just four. There are over forty times as many cores in this computer alone than there were at this year’s Quakecon).
Typically, supercomputers are measured by the number of arithmetic operations they can do each second, called FLOPS (floating-point operations per second). At its peak performance, in a single second, Stampede can do seven quadrillion of these operations. 7,000,000,000,000,000 — that’s 7 with fifteen zeroes after it.
Power to the People
Now, I know what you’re thinking. “I don’t pay extra tax on my vape juice so some bigwigs at the University can play SimCity.” But that’s the best part: anybody can get on board with and use the TACC supercomputers!
Allocations of computer time are dealt out through a front-facing portal on which new accounts are insanely easy to create. You need to enter less information to use a $50 million research tool than you do to make a Netflix account. And once you’re in their system, using an actual real-life supercomputer is as easy as logging in with SSH and queueing up tasks.
Here’s the catch: with a trial account, you get 1,000 computation hours up front, but beyond that you have to prove you’re actually doing important stuff. Any more hours require you to submit an Allocation Request detailing what you plan on doing with the machine. I’ll update back here if I ever get a response on my “Brute Force a Better Ending to Lost” proposal.
Divide and Conquer
Stampede’s full potential can’t be unlocked with just any old code, however. Processes that fully take advantage of TACC’s supercomputers have to be highly parallelized, meaning they’re crafted in a way that splits the work across potentially thousands of cores.
Example time: imagine you’re stocking up on groceries for the week, but you have just so many that the cashier’s going to take forever to ring them all up. Consult the following expertly-drawn diagram:
Instead, you could call up your friends and have them each take an item to a different cashier! You just saved a lot of time and built stronger friendships by taking a parallel approach.
Instead of items at a cash register, in parallel computing we’re queueing up separate sequences of instructions to run on different processors.
Pretty simple, right? Wrong. Programming for distributed systems like Stampede will give you a ton of headaches. If multiple processes need to update some shared data set, how do they do this while keeping this information consistent across the whole program? What should happen if one of the cores crashes?
Fear not. The kind Longhorns at TACC are dedicated to getting your important programs in shape for the big leagues. Not only has TACC stuffed their user portal with instructional guides, but they offer a number of live supercomputer training sessions, both in person and over webcasts. When was the last time you got to tell someone you’re taking supercomputer lessons?
A Most Triumphant Experience
My absolute favorite scene in Bill and Ted’s Excellent Adventure is when Keanu Reeves and his bud take Beethoven, Ghengis Khan, Joan of Arc and company to the shopping mall, and all these great historical figures just get their mind blown by our modern conveniences.
The analogy doesn’t quite carry over, but throughout our visit I found my mind jumping back to Laplace and his contemporaries. Would his visions of a deterministic universe sound nearly as far-fetched if he had lived alongside computational behemoths like Stampede?
I doubt the guy would be rocking out at TACC like Beethoven did on eight synthesizers, but it’s crazy to me that only 200 years have passed, and we’re taking major strides to making his dream a reality. If you’re a techie who wants a one-way ticket to Futuretown, I implore you to look into TACC and other supercomputing opportunities.
The possibilities are, well… super.