Benchmarking tool for RPC Servers

Project structure:

• benchmarker: RPC Client on Typescript. In charge of setting up multiple clients and making requests
• rs-server: RPC Server on Rust. The rs-server will clone the rpc-rust repo because it's still not published as a package.
• ts-server: RPC Server on Node. tsc --init project with minimum needed stuff.
• api.proto: Where the protocol and service are defined, shared by the benchmarker, rs-server and ts-server.
• book-large-content.txt: Large file that both servers use to store a book into an in-memory database which contains this large text in its content field.

How to use it?

Make sure you have node, rust and cargo installed.

Prepare environment

To have all ready to make benchmarks, you must run:

make build-ts

make build-rs

After both commands finished you are ready.

Run the tool

make ab server={SERVER_TO_RUN} n={TOTAL_REQUEST_TO_DO} c={CONCURRENCY_LEVEL_FOR_REQUESTS} 

• server: Value could be rs for Rust and ts for Typescript (node).
• n : Refers to total requests to be done by the benchmarker. eg: 10000 (used for the presentation benchmarking)
• c : Refers to the concurrency level of the tool. eg: 100 (used for the presentation benchmarking)

After running the tool, a .csv file will be created with a the type of the metric measure and the elapsed time. ${type},${elapsed_time}, there are two different types, one WHOLECONNECTION refers to the entire client connection and other one GETLARGEBOOKREQUEST that refers to just the request that fetches the book with a content field with 1MB.

The tool also throws an output with some stats similar to Apache Benchmarking -> mean, min, max, and percentiles (percentiles are just for the GetBook request). This output shows:

• Handle Client time metric: refers to the elapsed time for the entire client connection
• Rpc Client metric: refers to the elapsed time for the client creation and the connection is established with server.
• Rpc Port metric: refers to the elapsed time for the port creation, that it's also a request.
• Request time metric: refers to the elapsed time for the GetBook request (the same as GETLARGEBOOKREQUEST)
• percentiles metric: refers to the elapsed time percentiles for the GetBook request

Output example

Total test duration: {NUMBER}
Request / Second: {NUMBER}
Handle Client time (mean): {NUMBER}
Rpc Client (mean):  {NUMBER}
Rpc Port (mean):  {NUMBER}
Request time (mean):  {NUMBER}
Handle Client time (min):  {NUMBER}
Rpc Client (min):  {NUMBER}
Rpc Port (min):  {NUMBER}
Request time (min):  {NUMBER}
Handle Client time (max):  {NUMBER}
Rpc Client (max):  {NUMBER}
Rpc Port (max):  {NUMBER}
Request time (max):  {NUMBER}
Percentiles:  { '50': {NUMBER}, '75': {NUMBER}, '85': {NUMBER}, '95': {NUMBER}, '98': {NUMBER}, '100': {NUMBER} }

TODO:

• Adds a benchmarker-rs with the RpcClient written in Rust
• Adds C# implementation for benchmarking