SLA Concepts in System Architecture

Service Level Management & Capacity Planning In the world of Site Reliability Engineering (SRE), managing system performance isn't just about technical monitoring—it's about aligning technical reality with business promises. 1. The Hierarchy of Reliability: SLI, SLO, and SLA These three concepts form a hierarchical relationship. Think of it as a funnel: the SLI provides the data, the SLO sets the target, and the SLA makes the promise. SLI (Service Level Indicator): The metric . This is the actual measurement of a service's performance. It answers: "What are we measuring?" Examples: Request latency, error rate, throughput, or availability percentage. SLO (Service Level Objective): The target . This is the internal goal set for an SLI over a specific period. It answers: "How good is good enough?" Example: "99. 9% of requests will be completed in under 200ms over a rolling 30-day window. " SLA (Service Level Agreement): The contract . This is a formal, often legal, agreement between the service provider and the customer. If the SLOs are consistently missed, the SLA defines the consequences (e. g., service credits or financial penalties). It answers: "What is the business consequence if we fail?" 2. Capacity Planning Capacity planning is the proactive process of determining the resources (CPU, RAM, Storage, Network) your system needs to handle future demand. It is the bridge between "the system is working now" and "the system will work next year. " Why it matters: If you don't plan, your system will crash during peak traffic, or you will waste money on massive over-provisioning. The Process: Demand Forecasting: Analyzing historical data (Daily Active Users, Queries Per Second) to predict future growth. Bottleneck Identification: Finding the "weakest link" (e. g., a slow database query or a memory-intensive service) that limits your scale. Strategy Formulation: Deciding between Vertical Scaling (bigger servers) or Horizontal Scaling (more servers). Key Metrics for Estimation: QPS (Queries Per Second): The load on your application layer. Storage Growth: How much data your users generate per day. Network Throughput: The bandwidth required to serve your traffic. Reliability & Planning Reference Matrix Concept Nature Who cares? Primary Goal SLI Measured Data Engineering Team Quantify performance. SLO Internal Goal Engineering/Product Define acceptable reliability. SLA Legal Contract Legal/Customer Define business accountability. Capacity Resource Strategy Infrastructure/SRE Ensure sustainability & growth.