About
As part of Deno's continuous integration and testing pipeline we measure the performance of certain key metrics of the runtime. You can view these benchmarks here.
You are currently viewing data for the most recent commits to the masterbranch. You can also view all commits.
Runtime Metrics
In this section we measure various metrics of the following scripts:
- cold_relative_import
- text_decoder
- error_001
- cold_hello
- workers_round_robin
- relative_import
- workers_startup
- hello
Execution time
Log scale. This shows how much time total it takes to run a script. For deno to execute typescript, it must first compile it to JS. A warm startup is when deno has a cached JS output already, so it should be fast because it bypasses the TS compiler. A cold startup is when deno must compile from scratch.
Syscall count
How many total syscalls are performed when executing a given script. Smaller is better.
TypeScript Performance
Type Checking
In both cases, std/examples/chat/server_test.ts is cached by Deno. The workload contains 20 unique TypeScript modules. With check a full TypeScript type check is performed, while no_check uses the --no-check flag to skip a full type check.
I/O
Show normalized benchmarks
HTTP Server Throughput
Tests HTTP server performance. 10 keep-alive connections do as many hello-world requests as possible. Bigger is better.
- deno_tcp is a fake http server that doesn't parse HTTP. It is comparable to node_tcp
- deno_http is a web server written in TypeScript. It is comparable to node_http
- deno_core_single and deno_core_multi are two versions of a minimal fake HTTP server. It blindly reads and writes fixed HTTP packets. It is comparable to deno_tcp and node_tcp. This is a standalone executable that uses the deno rust crate. The code is in http_bench.rs and http_bench.js. single uses tokio::runtime::Builder::basic_scheduler and multi uses tokio::runtime::Builder::threaded_scheduler.
- hyper is a Rust HTTP server and represents an upper bound.
HTTP Latency
Max latency during the same test used above for requests/second. Smaller is better. Log scale.
HTTP Proxy Throughput
Tests proxy performance. 10 keep-alive connections do as many hello-world requests as possible. Bigger is better.
- deno_proxy_tcp is a fake tcp proxy server that doesn't parse HTTP. It is comparable to node_proxy_tcp
- deno_proxy is an HTTP proxy server written in TypeScript. It is comparable to node_proxy
- hyper is a Rust HTTP server used as the origin for the proxy tests.
Throughput
Log scale. Time it takes to pipe a certain amount of data through Deno. echo_server.ts and cat.ts. Smaller is better.
