SC072
Me and My Catapult
“I think of my Catapult as an origin, not a destination."
“I could think of it as a destination, as a destination for legacy Linux/MPI applications. These first-generation codes, including some of my own, started their lives in the sequential computing era. Then they were ported to MPI enabling them to run on the small clusters we had five years ago. The good ones scaled successfully to dozens of processors, but rarely as far as a hundred. So my 72-processor Catapult provides a good match to their scalability limits."
“But that is not how I use my Catapult. I use it to craft applications that will scale to thousands of processors. Is that overly egotistical of me? Not at all. In the years to come, all serious scientific computers will have thousands of processors. I want my work to thrive in this new world, not fall by the wayside."
"72 processors is the perfect launching pad for this work. The Catapult is inexpensive enough that I have my own right next to my desk; I can experiment to my heart’s content without having to sign up for processors at some distant data center. Which I often feel uncomfortable doing when I just want to try out an idea. And 72 processors is parallel enough to demand truly scalable thinking."
"There is a mental phase transition that comes along even before the threshold of 72 processors. It comes when it is no longer feasible to know, or care about, what each individual processor is doing. Once I shift from my old way of thinking about individual processors to thinking at the level of the ensemble, I am thinking in ways that work all the way up to thousands of processors."
“And thousands of processors is the destination that me and my Catapult are focused on. How about you?”
