Brussels / 4 & 5 February 2017

schedule

Scientific Computing on JRuby


I worked on “Port NMatrix to JRuby” in the context of GSoC 2016. The performance of NMatrix-JRuby is outstanding even without making use of JRuby threading capabilities. Also, I am working on ArrayFire gem that helps in easy GPU computation and is 1e4 to 1e7 times faster than NMatrix gem

JRuby is known for performance. NMatrix, a linear algebra library, now supports JRuby. All SciRuby gems that depend on NMatrix can now be run on JRuby. A scientific library must be highly efficient as the programs can be memory intensive. The programs must be fast at the same time. This calls for optimization.

Suppose, you are running a Ruby program and you need to make it faster; you use threads, yet you are not happy with the results! Just chain the Ruby methods to Java methods. Using Java method can improve the speed, around 1000 times when compared to using Ruby method. Not only speed, it can help you save a lot of RAM, around 10 times. JRuby’s Garbage Collection can sometimes slow down your program if you are not careful with handling a large amount of data.

I would also like to share about ArrayFire gem that would be soon available for MRI that can be used for high-performance GPU computing using OpenCL/CUDA.

Speakers

Photo of Prasun Anand Prasun Anand

Links