Ersel Hengirmen: Binary Search is Everywhere!!!
My name is Ersel Hengirmen. This year, I will be remotely participating in my project, “Chitchat, Gossip & Chatter – How to efficiently deal with communication,” at JSC-Jülich Supercomputing Centre with Ignacio Encinas Rubio. It is an amazing opportunity for every one of us. I am pretty excited to be a part of Prace Summer of HPC and more than ready to start my project, even though it would have been much more fascinating to be there and see the supercomputer up close.
Why am I here
I was born in Gaziantep/Turkey, and this September, I will be a fourth-year computer engineering student at Middle East Technical University(METU). I have always got fascinated by even the simplest of algorithms. I fell in love with computer science when I was implementing a basic program that plays the “guess the number” game. In the “guess the number” game, one player tries to find the picked number that must be between 1 to 1000 in at most ten tries. The guaranteed approach is to halve the remaining numbers constantly. It was a fascinating game I liked to play as a kid, but it turns out it was a fairly known algorithm known as binary search.
As for parallel programming, it was the genius that was behind the broadcast operation that made me fall in love this time. The algorithm used the theory behind binary search in reverse; by using the processes that received data to multiply sent data at every iteration. The weird part was that it didn’t matter if it was a ring, a hypercube, or any mesh topologies; the method always gave the same log(p) complexity.
As another weird anecdote: To be accepted to the Prace Summer of HPC we were given a task of finding error rates in pi calculation. To solve it I used a reversed version of binary search to find the interval of the solution and then used binary search to find the real solution. So binary search helped me get accepted to here.
Chitchat, Gossip & Chatter – How to efficiently deal with communication
Out of all the 22 projects, Chitchat, Gossip & Chatter was the one that spoke to me. In my college classes, we generally focused on the theoretical side and how some functions like broadcast and reduce work in the background. I chose Chitchat, Gossip & Chatter to put this theoretical knowledge to the test. In the following months, I plan to use and increase my knowledge in the theoretical territory while improving my skill in implementing this knowledge simultaneously, thus improving both aspects.
The project is about optimization. We will start with scaling up the Fast Multiple Method(FMM) for a large number of nodes to create a model that works better than naive point-to-point approaches. After this point, we will move according to our results. One of the important problems will be minimizing the overhead of communication, which can impede the speedup of the solver as the problem scales up. I believe that we will show you something we can be proud of at the end of this project. To learn more about it you can check the link here.
Who should apply
If you are interested in HPC, apply; that’s really it. It would be great if you knew about the subjects, but even if you don’t, the classes and resources our excellent instructors and mentors give you will be much more than enough. Just come here, look at the projects, find one that spokes to your heart, apply and see the fantastic world of HPC where everything you know can be done better. This field needs brilliant people like you, so come on in.