Blog

Technical blog posts.

1/3/26 Don't mount disk on EC2

ec2 disk mount

12/29/25 VLLM error and network configs

vllm error network configs

12/26/25 Roofline analysis for prefill and decoding

llm inference roofline analysis

12/24/25 comparison between TP and PP in llm inference

TP PP llm inference

12/2/25 My Way of Understanding vLLM V1 Scheduling Algorithm

vLLM scheduling memory management

11/22/25 GPU Communication Architecture: NCCL, NVSHMEM, and NVLink

GPU networking NCCL NVSHMEM NVLink

11/21/25 Memory bandwidth and latency for KV load

kv cache memory bandwidth latency

8/25/25 Kubernetes Scalability

k8s scalability etcd

8/21/25 How low disks space breaks envoy wasm cache and eventually makes http request routing fail (AIBrix LLM inference infra)

failure debugging k8s aibrix envoy wasm

7/12/25 Analysis on Various Overload Control Systems and their Limitaions

overload control system research

6/15/25 Debugging Forever Terminating Pods in Kubernetes

k8s debugging failure disk pressure

6/14/25 Deploying modified server without build, push, pull, and restart

flask k8s

5/18/25 Debugging HTTP Connection Overhead

failure performance debugging gateway aibrix

6/5/24 Debugging istiod failure: 'why is it so hard to find out disk pressure?'

k8s istio debugging istio ingress failure disk pressure

5/19/24 DCQCN, RDMA, CXL

networking notes

5/14/24 K8S cheat sheet

k8s notes

5/14/24 Intuition behind NVIDIA GPU architecture and CUDA programming model

GPU CUDA FLOPS