15. Horizontal Pod Auto-Scaling (HPA) for LLM Services based on TTFT (Time to First Token) & Queue Depth • AI Infrastructure, GPU Serving & MLOps (CUDA, Triton, TGI) • Rishabh Ki Kaksha

AI Infrastructure, GPU Serving & MLOps (CUDA, Triton, TGI) › 15. Horizontal Pod Auto-Scaling (HPA) for LLM Services based on TTFT (Time to First Token) & Queue Depth

50 mins

← Prev Next →

Course Syllabus

Module 1: GPU Architecture, CUDA Internals & Hardware Hierarchy

01. NVIDIA GPU Architecture: Streaming Multiprocessors (SMs), Tensor Cores, CUDA Cores & HBM3 Memory Bandwidth 02. GPU Memory Hierarchy: Registers, Shared Memory (SRAM), L1/L2 Caches & High-Bandwidth VRAM 03. CUDA Programming Basics: Writing Custom GPU Kernels in C++/CUDA & Triton Python DSL

Part 1: GPU Acceleration, Serving & Production Guardrails

Module 2: High-Performance LLM Serving Engines (vLLM, TGI, Triton)

04. The KV Cache Bottleneck: Memory Footprint Calculations & PagedAttention Virtual Memory Architecture 05. Dynamic & Continuous Batching (Iteration-Level Scheduling) vs Static Batching in LLM Serving 06. Production Benchmark: vLLM vs HuggingFace TGI (Text Generation Inference) vs NVIDIA Triton Inference Server 07. Speculative Decoding & Medusa Multi-Head Decoding for 2-3x Faster Token Generation

Module 3: Multi-GPU & Distributed Inference: Tensor & Pipeline Parallelism

08. Distributed AI Parallelism Paradigms: Data Parallelism (DP), Tensor Parallelism (TP), Pipeline Parallelism (PP) 09. Tensor Parallelism in Action: Megatron-LM Matrix Splitting across NVLink & NVSwitch Networks 10. DeepSpeed & PyTorch FSDP (Fully Sharded Data Parallel): Zero Redundancy Optimizer (ZeRO-1, ZeRO-2, ZeRO-3)

Module 4: MLOps Pipelines: Model Registry, Experiment Tracking (MLflow/W&B)

11. MLOps Foundations: Model Versioning, Artifact Registries & Lineage Tracking with MLflow and Weights & Biases 12. Automated Continuous Training Pipelines with Kubeflow Pipelines & GitHub Actions CI/CD for AI Models 13. Feature Stores in Production (Feast): Real-Time Online Feature Retrieval vs Offline Batch Processing

Module 5: Kubernetes GPU Orchestration, Auto-Scaling & Continuous Observability

14. Kubernetes for AI Workloads: NVIDIA GPU Operator, Device Plugins & Time-Slicing vs Multi-Instance GPU (MIG) 15. Horizontal Pod Auto-Scaling (HPA) for LLM Services based on TTFT (Time to First Token) & Queue Depth 16. Serverless GPU Architecture: Ray Serve, Modal & RunPod for Auto-Scaling AI APIs with Zero Cold Starts 17. Observability for GenAI: OpenLLMetry, Arize Phoenix, Langfuse & Prometheus GPU Utilization Metrics 18. Enterprise Cloud Cost Optimization: Spot GPU Instances, Multi-Cloud Orchestration & Slicing VRAM Costs

Interactive Lesson Studio

# 15. Horizontal Pod Auto-Scaling (HPA) for LLM Services based on TTFT (Time to First Token) & Queue Depth

> AI Infrastructure & MLOps Notes by Rishabh - Topic 15

---

### 1. Why CPU/Memory HPA Fails for LLM Serving ⚠️

LLM model load hote hi VRAM 90% allocate kar leta hai (Static KV Cache pre-allocation) $\rightarrow$ Standard Memory-based HPA **hamesha 100% false alarm scale trigger karegi!**

---

### 2. Custom Metrics Auto-Scaling with KEDA & Prometheus 📈

LLM Pods ko scale karne ke liye **True Operational Metrics** track kiye jaate hain:
1. **vLLM Waiting Queue Depth (`vllm:num_requests_waiting`):** Agar queue me 10 se zyada requests pending hon $\rightarrow$ Scale Up GPU Pods!
2. **Time to First Token (TTFT P95 Latency):** TTFT $> 500\text{ms}$ exceed hone par auto-scale trigger.
3. **KV Cache Usage Percentage (`vllm:gpu_cache_usage_factor`):** $> 85\%$ triggers scale out.

---

### 3. Interview Ke 3 Sawal Is Topic Se

1. **KEDA (Kubernetes Event-Driven Autoscaling) Prometheus metrics se HPA kaise drive karta as ScaledObject?** -> Prometheus query threshold define karke directly Kubernetes Deployment replicas scale karna.
2. **Scale-Down Stabilization Window (Cool-down period) kyu zaroori hai?** -> Flapping prevent karne ke liye (GPU instances ko turant terminate na karna taaki agle minute traffic spike par dobara cold boot penalty na lage).
3. **GPU Node Cold Start delay (3-5 minutes) kaise mitigate kiya jata hai?** -> Pre-warmed base images daemonsets, fast NVMe model caching, aur buffer over-provisioning pods configure karke.

---
**Next Topic:** 16. Serverless GPU Architecture: Ray Serve, Modal & RunPod for Auto-Scaling AI APIs with Zero Cold Starts

Loading lesson content...

Python 3.12

Full IDE

Live Execution Terminal

Click "Execute Live" to run code.
                

Lesson Mastery AI Quiz

AI Doubt Solver

CERTIFICATE OF COMPLETION

Rishabh Gehlot

PHP 8+ & Laravel 11 Master Track