Moving to the cloud swaps fixed hardware for resources that are elastic, shared and metered. That changes what "performance" means. A slow service might be short of CPU, throttled by a provider limit, waiting on a managed database, or perfectly fast but over-provisioned and wasting money. Cloud performance management covers both sides: the speed users experience and the efficiency of the resources behind it.
What it covers
- Application performance: response time, throughput and error rate per service and endpoint.
- Infrastructure: CPU, memory, disk and network on virtual machines, containers and managed services.
- Dependencies: databases, queues, caches and third-party APIs, which cause a large share of slowdowns.
- End-user experience: page load and API latency as users in different regions see them.
- Capacity and cost: how much of what you pay for is actually used, and how autoscaling behaves under load.
Why cloud performance is harder to manage
- Short-lived resources. Containers, serverless functions and autoscaled instances come and go, so tools that track fixed hosts lose sight of them.
- Provider limits. CPU credits on burstable instances, IOPS caps on volumes and API rate limits slow things down without any error on your side.
- Distributed requests. One request can touch many services, so the slow part isn't obvious without distributed tracing.
- Several consoles. Multi-cloud and hybrid setups split data across providers, each with its own metrics and naming.
Metrics that matter
- Latency percentiles (p50, p95, p99), not just averages, which hide the slow requests users complain about.
- Error rate and throughput per service.
- Saturation: CPU, memory, connection pools and queue depth.
- Apdex or a similar user satisfaction score.
- Utilization against provisioned capacity, to find over- and under-sized resources.
Good practices
- Set service level objectives on user-facing latency and errors, then tune infrastructure against them.
- Right-size from real utilization data rather than estimates.
- Compare performance before and after every deployment.
- Load test ahead of known traffic peaks.
Cloud performance management with Atatus
APM traces requests through services, databases and external calls and compares performance before and after each deployment. Infrastructure monitoring tracks hosts, processes and AWS and Azure metrics, and real user monitoring shows page load times and Core Web Vitals by country and browser. For a broader guide, read cloud monitoring.