Metrics to Measure

About

Load testing is not just about sending traffic — it’s about measuring the right system behaviors under stress. The success of a load test depends heavily on the metrics you capture, how you interpret them, and whether they align with your performance goals.

This section outlines the most important categories of metrics to track during load testing, across system layers and target types.

Category	Metric	What It Measures	Why It Matters	Where It's Relevant
User-Facing Performance	Response Time (Avg, P95, Max)	Time taken to serve a request (including network, processing, DB)	Indicates latency under load; high percentiles expose tail-end latency and bottlenecks	API, UI, Backend
	Throughput (TPS, RPS)	Requests/transactions processed per second	Shows system's processing capacity and how it scales	API, Messaging, Backend
	Error Rate (%)	% of failed requests (4xx, 5xx, timeouts, validation errors)	Affects user experience, SLA compliance, and system correctness	API, UI, Backend
	Concurrent Users / Sessions	Simultaneous users or sessions being handled	Determines concurrency limits and helps validate user load handling	All
	Request Completion Rate	Total successful vs attempted requests	Verifies system stability; low completion rate under load may indicate failures or dropped requests	All
Infrastructure & System	CPU Utilization (%)	CPU usage per service, container, or node	High CPU causes thread delays, processing lag, and degraded throughput	Backend, DB, Load Generator
	Memory Usage (Heap/Non-Heap)	RAM used during test; includes garbage-collected memory	Memory leaks or GC pressure can slow the app or crash it	Backend, JVM Apps
	GC Activity (JVM)	GC frequency, pause duration, memory reclaimed	Long GC pauses cause latency spikes and request timeout	JVM-Based Apps
	Disk I/O	Read/write latency, IOPS, queue time	Affects DB performance, logging, and caching behavior	DB, File-Intensive Apps
	Network I/O	Bytes sent/received, latency	Measures external call impact, CDN load, or backend communication lag	API, Integrations, Messaging
	Load Generator Usage	Resource utilization of load testing tool	Ensures tool doesn’t become the bottleneck and can handle simulated load	Load Generator
Database Behavior	Query Execution Time	Average and percentile durations for key queries	Identifies slow DB operations and poor indexing	API, Backend, DB
	Connection Pool Usage	Active, idle, waiting, and maxed-out connections	Pool saturation delays backend responses and increases error rates	Backend, DB
	Slow Query Count	Number of queries exceeding performance threshold	Highlights inefficient or blocking queries	DB
	Lock Contention / Deadlocks	Time spent waiting for locks or number of conflicts	Indicates concurrency issues or poor transaction design	DB
	Transaction Rate	Total commits/rollbacks per second	Reveals DB’s processing throughput, especially for write-heavy workflows	DB
Application Runtime	Active Thread Count	Number of currently executing threads	Reflects service load; rising threads may indicate pooling issues or blocking I/O	Backend, Microservices
	Thread Pool Utilization	Thread pool in-use vs available threads	Full pools mean requests queue or time out	Backend
	Request Queuing Time	Time a request waits before being processed	Helps detect queue saturation and slow backend response	Async Services, APIs
	Queue Depth (Messaging)	Length of processing queues (e.g., Kafka, RabbitMQ)	Backlogs show consumer underperformance or processing delay	Messaging, Event-driven Systems
	Retry / Fallback Counts	Circuit breaker or retry invocations	Excessive fallback shows downstream fragility or threshold breaches	Microservices
Functional Correctness	Business Flow Success Rate	% of transactions that completed the full functional flow	Validates that core business actions succeed under stress (e.g., order placed, payment complete)	All
	Data Integrity Issues	Duplicate, missing, or corrupted data under concurrency	Ensures correct handling of concurrent operations and transactions	DB, Backend
	Side-Effects Validity	Were all expected updates, logs, messages created?	Confirms completeness of side operations like audits, event pushes, etc.	Backend
Execution Details	Test Ramp-up Duration	Time taken to reach full user load	Helps observe system behavior as load increases gradually	All
	Steady-State Duration	How long the system was tested under full load	Important for soak or endurance testing, revealing gradual degradation	All
	Requests Sent vs Scheduled	How many requests were planned vs how many were executed	Detects tool-side issues or throttling limits on request injection	All
	Script / Assertion Failures	Errors from broken scripts or failed test assertions	Ensures script validity; separates tool errors from app behavior	All