Protobuf / JSON comparison

Test sample

The data used to extract the results below were generated once during the startup process of the Spring Boot application. It were generated 50 thousand people with "random data" and the following table exposes the data size that got sent through the wire on each request.

	gzipped size	non-gzipped size
protobuf	1.0mb	8.9mb
json	1.1mb	12.6mb

On a gzipped environment, protobuf payload was only 10% smaller, but on a non-gizzped env it was 30% smaller.

Front-end communicating with Java backend results

Deserialize (GET) from server

The following results were extracted from executing 15 requests per protocol on the test sample.

SpiderMonkey:

	gzipped time (avg)	non-gzipped time (avg)
protobuf	ms	620ms
json	ms	583ms

V8:

	gzipped time (avg)	non-gzipped time (avg)
protobuf	ms	524ms
json	ms	761ms

As we can see, when issuing requests from a JavaScript front-end app with protobuf.js it took us 72% of the time that it took on a json request. On a non-gzipped environment we were able to handle it even faster, it took only 49% of the time to handle a protobuf response compared to a json response.

Serialize (POSt) to server

The following results were extracted from executing 100 thousand post requests, per protocol, to the backend.

	avg ms / POST
protobuf	3.19527ms
json	3.31738ms

The results were quite similar. Protobuf post requests took only 4% less time than JSON post requests.

Two Java backends communicating results

The following results were extracted from executing 40 requests per protocol on the test sample.

	gzipped time (avg)	non-gzipped time (avg)
protobuf	234ms	146ms
json	701ms	590ms

For my surprise - and I believe everyone else's surprise - the Java backend test, communicating through a non-gzipped channel, got better results than the tests using gzip. Nevertheless, on both cases the Protobuf performance was much better. On the gzipped case, Protobuf took only 33% of the time that JSON took. On the non-gzipped case, Protobuf took 24% of the time that JSON took.

Advantages and disadvantages

Disadvantage: very few examples and poor documentation
Disadvantage: much smaller community (json tag: ~180.000; protobuf: ~200)
Advantage: 10 to 30% smaller payload
Advantage: 30 to 50% faster to be able to handle data

Generating protobuf files

The command below generates protobuf files for Java and also Javascript. You will ne to regenerate these files only if you change the ./src/main/resources/people.proto file or if you add new messages descriptions (new .proto files).

protoc --js_out=import_style=commonjs,binary:. \
       --java_out=./src/main/java/ \
       ./src/main/resources/people.proto

browserify ./src/main/resources/people_pb.js -o ./src/main/resources/static/bundle.js

brunokrebs/auth0-speed-test