On 9/19/2013 3:14 AM, Anton M wrote: > Shawn, I had swap file growing (up to 50-60%) and working while load tests > ran. Did you configure 'swapiness' on your Linux box (set it to 0 earlier, > maybe)? If not, my Windows OS could be cause of that difference.
The vm.swappiness sysctl setting is 1. I have used 0 as well. I don't want it to start swapping unless it *REALLY* needs to. The default of 60 is pretty aggressive. > I'm not sure if that's completely an issue about shareable memory or some > missing JVM configurations (I don't have anything special except -Xmx, -Xms > and -XX:MaxPermSize=512M) or some Solr memory leak. > I'd appreciate any thoughts on that. As I said before, I think that the memory reported as shareable is not actually allocated. It probably should be listed under virtual memory. Our app rarely does facets, and it typically sorts on one field, so I have absolutely no idea what's being measured in the 11g of shared memory for the solr process. I was present for a conversation between Lucene committers on IRC where they seemed to be discussing this issue, and it sounded like it is a side effect of using MMap in a particular way. It sounded like they didn't want to change the way its used, because it was the correct way of using it. The conversation went way over my head for the most part. Thanks, Shawn
