Posts

A deep dive into how STL containers are implemented and what that means for performance — vector, the tree-backed set/map, hash containers, and when to reach for each.

A short recipe for running a Jupyter notebook inside a remote Docker container and reaching it from your laptop’s browser or VS Code via two port hops.

A cheat sheet of the Linux tools I reach for when diagnosing disk, network, memory, CPU, and GPU — plus the one gotcha about how nvidia-smi’s utilization number actually works.

Practical knobs for getting more out of iperf — parallel streams, larger TCP windows, longer test durations, disabling Nagle’s algorithm, and when to switch to UDP or SCTP.

Notes from the 2022 ACM HPC Summer School in Barcelona — RISC-V for ML, vector extensions, distributed data analysis, and energy efficiency in HPC.

NFS vs Lustre — the architectural trade-offs (stateless vs stateful, in-band vs out-of-band I/O) and which one to pick for which workload.

Slides from a talk I gave at the SenseTime-NTU lab on accelerating distributed deep-learning training under data parallelism.

Move semantics let the compiler swap expensive copies for cheap moves. A walkthrough of lvalues vs rvalues, when moves actually fire, and why they matter for performance.

Stack vs heap in C++ — why the stack is faster, how each is managed by the OS, and what changes when threads and processes enter the picture.

A self-contained shell script for building tmux from source into your home directory — useful on shared machines where you don’t have sudo access.