Build a From-Scratch LLM Serving Scheduler with Chunked Prefill, Prefix Caching, and Preemption
Learn to build a lightweight LLM scheduler that demonstrates real systems skills, from architecture to production extensions.
Scheduling
Mastering Zephyr RTOS Scheduling for Multi-Sensor Edge Nodes: Real-Time Performance Tuning
Real-time edge nodes demand deterministic scheduling. This post shows how to tune Zephyr RTOS thread priorities, interrupt handling, and memory for multi-sensor workloads.
Tuning Linux Kernel Network Buffers and Scheduling Policies for High‑Performance Networking
Table of Contents Introduction Why Kernel‑Level Tuning Matters Anatomy of the Linux Network Stack 3.1 Socket Buffers (sk_buff) 3.2 Ring Buffers & NIC Queues Core Network Buffer Parameters 4.1 /proc/sys/net/core/* 4.2 /proc/sys/net/ipv4/* Practical Buffer Tuning Walk‑through 5.1 Baseline Measurement 5.2 Increasing Socket Memory Limits 5.3 Adjusting NIC Ring Sizes 5.4 Enabling Zero‑Copy and GRO/LRO Scheduling Policies in the Kernel 6.1 Completely Fair Scheduler (CFS) 6.2 Real‑Time Policies (SCHED_FIFO, SCHED_RR, SCHED_DEADLINE) 6.3 Network‑Specific Scheduling (qdisc, tc) CPU Affinity, IRQ Balancing, and NUMA Considerations Putting It All Together: A Real‑World Example Monitoring, Validation, and Troubleshooting Conclusion Resources Introduction Modern data‑center workloads, high‑frequency trading platforms, and large‑scale content delivery networks demand sub‑microsecond latency and multi‑gigabit throughput. While application‑level optimizations (e.g., async I/O, connection pooling) are essential, the Linux kernel remains the decisive factor that ultimately caps performance. ...