As a data point: I'm managing (or more watching and daily sacrificing a goat to) a JVM using Hotspot parallel GC with a 14GB heap and 4.5GB young generation. It is randomly taking 30-90 seconds to do a minor GC due to scheduled quartz job generates large XML document POST request that I suspect is around 4GB in size. There is no VM memory pressure or swapping, as the VM has 16GB of RAM.
As a general rule I've found young generations of more than about 2GB and JVM heap sizes of more than 8GB (i.e. large old generations) really do not perform well with the Hotspot parallel GC when there is a mix of small requests creating a lot of short-lived small objects and then a large slow request hits the young generation. As I can't yet reproduce this problem in a test environment I'm not sure whether the G1 collector might perform better. Naturally, Azul System's Zing JVM would likely be a simple solution to dealing with this, but as this is a retail platform under a change freeze, so my goal is to keep it alive until Christmas... Ho ho ho, Gary -- You received this message because you are subscribed to the Google Groups "mechanical-sympathy" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
