PD-Disaggregated Inference Deployment with sgl-project/rbg
Use RoleBasedGroup to run SGLang Prefill/Decode disaggregation as one operable unit: routing, startup dependencies, multi-node tensor parallelism, KV transfer, and coordinated rollouts.
CFN Cloud
Kubernetes, GPU, and AI Infrastructure Guides
Practical tools
Use the tools while you read
Browse by topic
Popular topics
Editors' picks
Featured reads
Kubernetes Probes: Liveness, Readiness, and Startup
Use better Kubernetes probes by choosing the right signal, tuning thresholds, and avoiding false restarts, traffic drops, and noisy rollouts.
Linux Memory Allocation: malloc, brk, mmap, and RSS
Use a reproducible program, /proc, and strace to inspect how glibc malloc uses brk, mmap, and madvise, and why RSS may stay high after free.
Kubernetes Ephemeral Containers: Safe Production Debugging
Safely inspect a live Pod without baking debugging tools into production images.
Kubernetes NetworkPolicy: Default Deny, DNS, and Safe Rollout
A concrete rollout path for Kubernetes NetworkPolicy: start with default deny, whitelist DNS and key dependencies, and avoid breaking production traffic.
Startup GPU Costs: Serverless, Dedicated, and Hybrid Capacity
Calculate Serverless versus dedicated GPU break-even using effective GPU hours, cold starts, idle capacity, operations, and SLOs, then plan hybrid capacity.
Read by topic
Topic guides
Kubernetes
A curated reading track for Kubernetes.
Systems
A curated reading track for Systems.
GPU
A curated reading track for GPU.
Latest posts
Recent writing
PD-Disaggregated Inference Deployment with sgl-project/rbg
Use RoleBasedGroup to run SGLang Prefill/Decode disaggregation as one operable unit: routing, startup dependencies, multi-node tensor parallelism, KV transfer, and coordinated rollouts.
Kubernetes Probes: Liveness, Readiness, and Startup
Use better Kubernetes probes by choosing the right signal, tuning thresholds, and avoiding false restarts, traffic drops, and noisy rollouts.
Linux Memory Allocation: malloc, brk, mmap, and RSS
Use a reproducible program, /proc, and strace to inspect how glibc malloc uses brk, mmap, and madvise, and why RSS may stay high after free.
Kubernetes Ephemeral Containers: Safe Production Debugging
Safely inspect a live Pod without baking debugging tools into production images.
Kubernetes NetworkPolicy: Default Deny, DNS, and Safe Rollout
A concrete rollout path for Kubernetes NetworkPolicy: start with default deny, whitelist DNS and key dependencies, and avoid breaking production traffic.
Startup GPU Costs: Serverless, Dedicated, and Hybrid Capacity
Calculate Serverless versus dedicated GPU break-even using effective GPU hours, cold starts, idle capacity, operations, and SLOs, then plan hybrid capacity.
Kubernetes Rollout Safety: PDB, Surge, and Unavailable Settings
Combine Deployment rollingUpdate settings with PodDisruptionBudgets to keep availability during upgrades and node maintenance.
LWS PD Disaggregation for vLLM and SGLang
Deploy SGLang and vLLM Prefill/Decode roles with LWS DisaggregatedSet, NIXL startup commands, validation steps, and production safeguards.
GPU Overprovisioning: Time-Slicing, MIG, and Backfill
Compare queue backfill, NVIDIA Time-Slicing, MIG, and runtime sharing with Kubernetes configuration, validation, isolation, and rollback steps.
Kubernetes Requests and Limits: Scheduling, Throttling, OOM
How CPU/memory requests and limits actually affect scheduling, throttling, OOMKills, and autoscaling.
Linux x86-64 Function Calls: Stack Frames, ABI, and GDB
Use a reproducible C program and GDB session to inspect SysV AMD64 arguments, call/ret, stack frames, unwinding, PIE addresses, and optimization.
Kubernetes RBAC Least Privilege: Roles and Bindings
Learn concrete Kubernetes RBAC least-privilege patterns, how to reduce overbroad permissions, and which checks catch risky role bindings before incidents.
Linux Dynamic Linking: GOT, PLT, Lazy Binding, and dlopen
A walkthrough into how Linux dynamically links shared libraries at runtime — PIC, GOT, PLT, lazy binding, gdb tracing, and the real cost of -fPIC.
Linux Stack Overflow Protection: Canaries, NX, ASLR, PIE, and RELRO
Build and inspect a reproducible Linux binary to understand Stack Canaries, FORTIFY, NX, Stack Clash protection, ASLR, PIE, RELRO, and CET.
Linux Heap Memory: Arenas, Chunks, Bins, and Tcache
Use a reproducible glibc malloc experiment to separate real leaks, allocator caching, fragmentation, and RSS retention across Arenas, Chunks, Bins, and Tcache.
OpenClaw Ecosystem: Compare Five Self-Hosted Agent Frameworks
A September 3, 2026 snapshot of five self-hosted agent frameworks, comparing deployment, extensions, security boundaries, and operational cost.
KAI-Scheduler vs HAMi: Scheduling Accounting and Runtime Enforcement
Compare KAI-Scheduler Reservation Pod accounting with HAMi CUDA API-level memory and compute enforcement, including their boundary versus hardware isolation such as MIG.
Kubernetes vs Docker vs OpenStack: Where Each Fits
Separate container packaging, Kubernetes orchestration, and OpenStack IaaS so teams choose the layer that matches their actual operating problem.
hetGPU: What Cross-Vendor GPU Binary Portability Requires
A review of the hetGPU research prototype that separates paper design, conceptual pseudocode, and current cross-vendor GPU engineering capability.
Linux cgroups: V1, V2, CPU Throttling, and OOM Debugging
Debug Linux cgroup V1 and V2 with cpu.stat, memory.events, PSI, and Kubernetes resource settings to explain CPU throttling, reclaim, and OOM kills.
gpu-manager Architecture: Device Plugin, vCUDA, and Migration
Review tkestack/gpu-manager device registration, vCUDA resources, library injection, topology allocation, CUDA limits, and migration choices.
ELF Files: Sections, Segments, Relocations, and Dynamic Linking
Understand ELF files from sections and segments to relocations and dynamic linking, with concrete examples for debugging Linux binaries and loader issues.
Kubernetes Troubleshooting: Pending, CrashLoopBackOff, and Traffic
A concrete Kubernetes troubleshooting playbook for Pending Pods, CrashLoopBackOff, readiness failures, networking issues, and node-level problems.
Kubernetes Autoscaling: HPA, VPA, and Cluster Autoscaler Boundaries
How to make autoscaling predictable: right requests, sane HPA behavior, VPA recommendations, and capacity-aware cluster scaling.
Kubernetes Probes: Liveness, Readiness, and Startup Protection
Learn how liveness, readiness, and startup probes work in Kubernetes, what each one should check, and how to avoid restart loops and false failures.
Deploying MySQL with Helm: Chart Defaults and Operational Boundaries
Use Helm to deploy a MySQL cluster on Kubernetes while understanding chart defaults, persistence, networking, and production tradeoffs.
kubectl port-forward: Debugging Boundaries and Failure Signals
Learn how kubectl port-forward works, when to use it for debugging, and how it differs from Services, Ingress, and production traffic paths.
MySQL Source and Replicas on Kubernetes: Status, Routing, and Failover
Understand MySQL source-replica topology on Kubernetes and use MySQL 8.4 commands to verify replication status, routing, and failover.
Kubernetes Headless Service: Pod Identity, DNS, and Stateful Apps
Understand when to use a headless Service in Kubernetes, how DNS works without a virtual IP, and why it matters for StatefulSets and peer discovery.
Kubernetes StatefulSet: Stable Identity, Ordered Rollouts, and Storage
Learn when to use a StatefulSet in Kubernetes, how stable Pod identity works, and why ordering and persistent storage matter.