Performance

Redis Cluster architecture diagram showing memory distribution across shards

Optimizing Redis Cluster Eviction: Predictable Latency Under Memory Pressure

Redis Cluster eviction under memory pressure is a common source of unpredictable latency spikes. This post breaks down the eviction mechanics, identifies the failure modes, and provides concrete strategies to keep p99 latency stable.

A schematic of memory-mapped file access across sharded model weights.

Building a Safetensors Schema Validator and mmap-Backed Tensor Reader for Sharded Model Weights

This project demonstrates how to validate safetensors schemas and read sharded model weights using memory-mapped files, showing practical systems expertise.

A conceptual illustration of a token sampling distribution

Building a Production-Grade Top‑k/Top‑p Sampler with NumPy

This tutorial shows how to build a high‑performance top‑k/top‑p token sampler with temperature scaling and per‑token log‑probability tracking using NumPy. It demonstrates real systems skills relevant to ML inference roles.

eBPF bytecode attaching to kernel functions for performance tracing

Deep Dive into eBPF: Tracing Kernel Functions for Performance Analysis

A practical deep dive into eBPF kernel tracing, covering function hooking, map-based data collection, and real-world performance profiling patterns.

Illustration of a multi-tiered compiler pipeline over a polyglot runtime

Architecting High-Performance JIT Compilation in the GraalVM Polyglot Runtime

GraalVM’s tiered JIT adapts compilation based on runtime profiles. Polyglot programs see up to 3x speedups.