Liberica JDK Performance Edition

• CPU: Intel(R) Xeon(R) CPU X5675 @ 3.07GHz,
• Memory: 96GB,
• OS: Linux Ubuntu 22.04.1.

To measure GC evacuation pause affecting application latency, we tested G1GC (a default GC in Java 11) with the BigRamTester benchmark. The tests evaluated G1 Evacuation Pause (where live objects are copied from one region into another), when all application threads are stopped for the period of garbage collection.

For throughput evaluation, we utilized the SpecJBB benchmark, which measures throughput of an application with two metrics:

• max-jOPS represents the maximum transaction throughput of a system until failure;
• critical-jOPS is the mean stable transaction throughput in the long term.

• CPU: Intel(R) Xeon(R) CPU X5675 @ 3.07GHz,
• Memory: 96GB,
• OS: Linux Ubuntu 22.04.1.

When studying throughput and latency, we used two GC implementations: G1GC and ZGC. ZGC is a scalable low latency garbage collector absent in OpenJDK 8. We measured latency using the BigRamTester benchmark (1h run with 50 GB).

For throughput evaluation, we utilized the SpecJBB benchmark, which measures throughput of an application with two metrics:

• max-jOPS represents the maximum transaction throughput of a system until failure;
• critical-jOPS is the mean stable transaction throughput in the long term.

To measure the overall performance, we used the DaCapo benchmark, a set of real-world Java applications with different memory loads.

Finally, the study of compression/decompression speed was done using two datasets, Large Calgary Corpus and Silesia Corpus.