Relative Performance of Java's Garbage First (G1) Garbage Collector? -

Does anyone know about any performance standard of Java's first garbage (G1) garbage collector ("Old "Compared to GC)?

In the context of the time of GC breaks, Sun says that G1 is sometimes better and sometimes worse than CMS. While the G1 collector has been successful in limiting the time of total break, it is still only a soft real-time collector. In other words, it can not guarantee that this application will not affect the ability to complete the deadline of threads, all these times However, it can work within a well defined set range, which makes it ideal for a soft real-time system that needs to maintain high-throughput performance.

I want to see the real measures of throughput: Java's CMS (concurrent mark broom) and G1 (trash first) delay for the collectors.

The original scientific article (by Ditfords, Flood, Heller and Printesis) contains some details in the actual measures (in section 4).