02. DevOps about Plan

DevOps about Plan

---

Prerequisites

---

1. What is Plan in DevOps?

The Plan phase is where everything starts.

Before writing a single line of code, you define:

• What problem you are solving
• What the system should do
• How performance will be measured
• What constraints exist

Plan is the phase where requirements, constraints, and system goals are clearly defined before implementation begins.

2. Why Planning Matters (Especially for C++ Systems)

In high-performance systems (like C++ real-time processing), bad planning = expensive rework

❌ Without Planning

• Wrong architecture → rewrite needed
• Latency too high → redesign pipeline
• Memory issues → major refactor
• Not scalable → system collapse under load

✔ With Proper Planning

• Clear performance targets
• Predictable system behavior
• Easier optimization later
• Faster development cycle overall

Example

Let’s assume we are building:

A C++ image preprocessing pipeline for real-time industrial inspection

4. Step-by-Step Planning Breakdown

4-1. Define the Problem

What are we solving?

• Input: raw camera frames
• Output: processed images for downstream vision algorithms

4-2. Define Requirements

Functional Requirements

• Resize images
• Apply filtering (denoise, normalize)
• ROI extraction

Non-Functional Requirements (Critical)

• Latency: < 10 ms per frame
• Throughput: > 100 FPS
• Stability: 24/7 operation
• Accuracy: preprocessing must not degrade detection quality

In C++ systems, non-functional requirements are often more important

4-3. Define Constraints

• Hardware:

  • CPU only? GPU allowed?

• Memory limits:

  • Can we buffer frames?

• Environment:

  • Embedded system vs server

Example:

• CPU: 8 cores
• No GPU
• Limited memory (512MB)

4-4. Define System Architecture (High-Level)

Design the pipeline before coding:

Camera Input
   ↓
Frame Queue
   ↓
Preprocessing Stage (parallel)
   ↓
Output Queue
   ↓
Consumer (Vision Algorithm)

4-5. Identify Performance Bottlenecks Early

Ask:

• Where will time be spent?
• Where can parallelism be applied?

Example:

Stage	Risk
Image copy	memory bandwidth bottleneck
Filtering	CPU-heavy
Queue sync	lock contention

4-6. Choose Optimization Strategy

Based on planning:

• SIMD (AVX/SSE) for pixel operations
• Multithreading (per-frame parallelism)
• Cache-friendly memory layout (SoA vs AoS)

4-7. Define Metrics & Observability

Before coding, decide how to measure success:

• Latency per frame
• Throughput (FPS)
• CPU usage
• Memory usage

4-8. Define Build & Dependency Strategy

Even in planning, DevOps thinking starts:

• Build system → CMake
• Dependencies → vcpkg
• Target platform → Linux / Docker

5. Output of the Plan Phase

At the end of planning, you should have:

• Clear system requirements
• Defined architecture
• Performance targets
• Identified risks
• Tooling decisions

6. Common Mistakes

❌ Jumping into coding too early

❌ Ignoring non-functional requirements

❌ No performance targets

❌ No architecture definition