A GC cycle has a latency overhead. For critical applications, this can affect the desired throughput. With a carefully selected combination of multiple parameters, such as heap size, allocation units, GC cycles duration, region size, and various other parameters, you can compare the performance of your application across various GCs (including Epsilon).

However, overheads can also be induced by the system, which is not related to a GC. Executing your applications with multiple GCs will enable you to separate out GC-induced overheads from system overheads and select the best GC for your application.

By using a no-op GC such as Epsilon, you can filter out the GC-induced performance issues with OS/compiler issues.