Performance

What reviewers are scoring when they score performance, and what it does not cover.

Performance is one of the four dimensions a reviewer scores separately. It asks about the thing the customer paid for: does the machine do the work at the speed the listing implied.

What reviewers are asked about

• Whether the hardware behaves like the specification suggested.
• Whether performance is consistent, or varies with the time of day in the way heavily oversubscribed hosts do.
• Disk and network throughput in ordinary use.
• How the machine behaves under the load the reviewer actually put on it.

What it is not

Not a benchmark. Nobody runs a standard workload here. These are impressions from people using the machines for their own purposes, which is both the limitation and the value: a benchmark tells you about the benchmark, and a hundred customers tell you about the experience.

Not the same as the latency figure. Monitoring measures round-trip time to the provider's network from many places. That is network distance, not machine speed. A provider can have excellent latency and disappointing hardware, and the listing shows both because they are different questions.

Reading it

Performance scores are the most sensitive of the four to what the reviewer was doing. A host that is excellent for a small web application and poor for a database can pick up a wide spread of honest scores. A spread is therefore worth reading as information rather than as noise: the distribution is shown, not just the average.

The count matters more here than on the other dimensions. Three performance scores tell you about three workloads.

Where the objective figures are

For anything measured rather than reported, see Monitoring metrics. That section covers uptime, response time and incident history, which are independent of reviewer opinion entirely.