Benchmark for Image Steam to help determine ideal hardware configurations and load levels. Designed to support load testing of large configurations, where every simulated user has it's own dedicated thread, and the origin proxy has it's own dedicated process.
isteambHTTP mapping to your Image-Steam configuration, allowing this benchmark to act as origin.
- Install via
npm i -g image-steam-bench
- Run benchmark via
isteamb run http://localhost:8080/isteamb --port 12124, where the URL is pointing to the Image-Steam endpoint, and the port is where Image-Steam is configured to connect to for
12124) - Port for image-steam-bench to listen on (same port image-steam should be mapped back to).
webp) - Image format requested in every test.
origin optimized cached real-90 real-95) - One or more tests to run.
1.25) - Increase in mean response times before considered minimum safe load.
2.0) - Max load is determined by optimal TTFB multiplied by this value.
20000) - Minimum time (in ms) that a test must run before determinating. If
requestsis specified that will take priority.
requests- A fixed number of requests before resolving test(s), versus the default behavior of ending on
1) - Number of workers to start out with.
999) - Maximum number of workers allowed.
3) - Seconds before new workers are spawned.
0.2) - The rate at which workers are spawned (0.2 being +20% per spawn).
1) - Seconds between updates.
30) - Seconds represented on historical graphs.
isteamb.log) - Filename of activity log, or
Gradual increase in concurrency load until errors or timeouts begin to occur, starting with a concurrency of 1 which is a measure of performance instead of load.
- Original - Worst case performance where nothing is cached.
- Optimized - Optimized originals available, but image artifacts must still be created.
- Cached - Best case scenario, no image operations, pure cache throughput.
- Real 90/8/2 - Emulation of more realworld environment where 90% of hits are cached, and 8% use optimized originals to create final image artifact, and 2% hit original and create both optimized original and image artifact.
- Real 95/4/1 - Emulation of more realworld environment where 95% of hits are cached, and 4% use optimized originals to create final image artifact, and 1% hit original and create both optimized original and image artifact.
Calculated for each of the tests.
- Performance (50th/75th/90th TTFB ms) - Single concurrency score to demonstrate raw performance.
- Minimum Load (req/sec @ concurrency, 50th ms) - The level of load before per-request response times begin to creep up.
- Optimal Load (req/sec @ concurrency, 50th ms) - The level of load that is considered ideal maximum before throughput begins to drop.
A number of real-time data points are available from the dashboard.
- Throughput (per/sec) - Historical graph
- Errors (per/sec) / Concurrency - Historical graph
- Latency (ms) - Historical graph
- Test Progress (%) - Progress of all tests
- Scores - See
- Activity Log
rps- Requests per second.
cc- Concurrency, the number of tasks taking place at the same time. In the context of
isteamb, each task resides on its own worker thread.
Kb- Kilobit, or 1000 bits, or 125 bytes.
Mb- Magabit, or 1000 kilobits, or 125 kilobytes.
Gb- Kilobit, or 1000 megabits, or 125 megabytes.
/s- Per second.
TTFB- Time to first byte. The time it takes from initiating a task to receiving the first data of the response. This may also be referred to as latency or response time.
50th- The sorted 50th percentile of data, or the "mean".
75th- The sorted 75th percentile of data, or the 25th percentile highest data point.
90th- The sorted 90th percentile of data, or the 10th percentile highest data point.
A dedicated worker process, serving files from memory to emulate high throughput origin.
http://localhost:12124/isteamb/12mp.jpeg/*- 12 magapixel (4256x2832) asset
http://localhost:12124/isteamb/18mp.jpeg/*- 18 magapixel (5184x3456) asset
http://localhost:12124/isteamb/24mp.jpeg/*- 24 magapixel (6016x4016) asset
12mp.jpeg- Photo by NASA on Unsplash
18mp.jpeg- Photo by Justin Clark on Unsplash
24mp.jpeg- Photo by Casey Horner on Unsplash