Disabling swap entirely is usually the easiest fix, yes. On Mon, Sep 20, 2010 at 8:10 PM, Alaa Zubaidi <alaa.zuba...@pdf.com> wrote: > Thanks Peter, > I decreased the heap size, it did not help, however, it delayed the problem. > I noticed that its swapping, so, do you think that I should set windows to > Not to swap? > > Do you think its related to this issue? > https://issues.apache.org/jira/browse/CASSANDRA-1014 > > Thanks, > Alaa > > On 9/18/2010 7:04 AM, Peter Schuller wrote: >>> >>> I see a spike in heap memory usage on Node 2 where it goes from around 1G >>> to >>> 6GB (max) in less than an hour, and then goes our of memory. >>> There are some errors in the log file that are reported by other people, >>> but >>> I don't think that these errors are the reason, because it use to happen >>> even before I have seen them. >>> >>> Can someone help me understand whats going on?? >> >> Only partially: >> >> 54 seconds ParNew GC:s are wild and crazy: >> >> INFO [GC inspection] 2010-09-17 14:53:59,403 GCInspector.java (line >> 129) GC for ParNew: 54095 ms, 53297952 reclaimed leaving 4712568360 >> used; max is 6563430400 >> >> Is the machine swapping? >> >> I noticed there is hinted hand-off activity going on. Maybe that is a >> result of nodes dropping in and out due to swapping. In any case, you >> definitely don't want to have the machine swapping to death. I'm not >> sure what the best way is to avoid this on Windows, other than >> decreasing heap size. >> >> The repeated exceptions in your log aren't normal as far as I know. I >> don't remember, but IIRC the UTF-8 encoding issues can be caused by >> changing partitioner after inserting data (but I'm not sure). >> >> With respect to memory use, you don't seem to be inserting so much >> data for bloom filters and sstable index samples to be a problem. >> Memtable flushing could cause problems if they're happening too slowly >> (maybe plausible with swapping) - except that the stage statistics >> don't indicate there are lots of memtables in memory waiting to be >> flushed, so that shouldn't be it. >> >> Hinted handoff maybe, but I don't remember whether hinted handoff has >> the potential to accumulate data in RAM (no time to check now). >> >> Regardless, I'd recommend fixing any swapping issues you have before >> trying to draw conclusions about performance. And you don't want those >> exceptions in your logs. >> > > -- > Alaa Zubaidi > PDF Solutions, Inc. > 333 West San Carlos Street, Suite 700 > San Jose, CA 95110 USA > Tel: 408-283-5639 (or 408-280-7900 x5639) > fax: 408-938-6479 > email: alaa.zuba...@pdf.com > > >
-- Jonathan Ellis Project Chair, Apache Cassandra co-founder of Riptano, the source for professional Cassandra support http://riptano.com
