A benchmark of the following JSON-LD implementations:
- Titanium: Implements the JSON-LD 1.1 spec with a high coverage.
- Json-LD Java: Implements the JSON-LD 1.0 spec with a high coverage. Implements part of the JSON-LD 1.1 API (still low coverage).
The benchmarks use a subset of W3C compliance test suite for JSON-LD 1.1. In order to compare the 2 implementations, we need the subset of test entries which pass in both implementations. Because of the low ow coverage for Json-LD Java on Json-LD 1.1 spec this subset is sometimes small.
The default parameters are: 5 JVM warmup iterations and 10 measurement iterations:
sbt jmh:run
You can tweak the warmup iterations and the measurament iteration parameters:
sbt jmh:run -iw {warmup_iterations_number} -i {measurement_iterations_number}
The test suite can be found as a compress bundle in src/test/resources/json-ld-11.org.tgz from the original W3C compliance test suite.
However, tests can be updated (in case W3C modifies them) using the following command (Mac OS and Linux):
bash download_suite.shThe provided results are the ratio between the throughput obtained by Json-LD Java implementation / the one obtained by Titanium.
| Number of tests | Ratio (Json-LD Java / Titanium) | |
|---|---|---|
| compact | 89 | 7.62 |
| expand | 83 | 7.34 |
| flatten | 45 | 10.60 |
| frame | 42 | 17.13 |
| fromRdf | 30 | 5.73 |
| toRdf | 127 | 11.05 |
| Total | 416 | 9.911 |
from the benchmark results the Json-LD Java implementation is 10 times faster in average than Titanium.
This difference in performance can be caused by the greater complexity of the different algorithms in Json-LD 1.1 due to the introduction of many new features.
The fact that Titanium is a very recent library (at the time of writing, 03.12.2020) can also explain part of the difference in benchmark results, compared to a mature Json-LD Java.