In my previous post I traced RDMA AllReduce on SoftRoCE using eBPF. SoftRoCE runs RDMA over the Linux kernel network stack, so eBPF could attach probes and observe the data path directly. This post...

Most explanations of NCCL and InfiniBand stay at a high level. I wanted to understand what actually happens at the kernel level during GPU-to-GPU data transfer, so I built an RDMA ring allreduce fr...

Introduction In Scala functions are first-class values which means functions are just like other values in Scala. We can pass functions as arguments to other functions, stored the function in valu...

Introduction Scala model of expression evaluation is based on the principle of substitution model, in which variable names are replaced by the values they are bound to.The underlying idea is that ...

Now days I am spending a quite a bit of time with Python from doing hobby project to implementing algorithms and data structures to writing Apache Spark code. Python is one of my favorite language....

Background The Spark has bottleneck on the shuffling while running jobs with non-trivial number of mappers and reducer. There has been lots of improvement in recent release on shuffling like conso...

Introduction This is the project for the course Practical Machine Learning from the Coursera.This projects uses the accelerometer measurement of 6 people over the time. The data contains the accel...