That sounds a little odd, may help if you post the yaml settings, the tpstats and the log lines that look something like this...
INFO [ROW-MUTATION-STAGE:5] 2010-09-04 15:43:49,402 ColumnFamilyStore.java (line 790) Enqueuing flush of memtable-sup...@1754565178(80208 bytes, 2304 operations) Also wondering why you chose such high memtable thresholds and how that's working. Do you expect to trigger the operation count or the throughput count in normal processing? Have you seen the additional guidance on mem table tuning here http://wiki.apache.org/cassandra/MemtableThresholds Aaron On 7 Sep 2010, at 18:33, Mubarak Seyed wrote: > I have a 8 nodes cluster, MemtableThreshold is 2 GB/CF, MemtableObjectsCount > is 1.2, heap (min/max) is 30 GB and only 4 ColumnFamilies > > It appears from system.log that flush happens for every < 50 operations (read > or write) and compaction is happening very frequently and i could see lots of > sstable is getting created (with smaller size). For a just 1000 inserts, i > could see around 20 sstables. > > When i change the MemtableThreshold to 1 GB/CF, everything works as desired. > > Any idea, what could be the problem when i specify MemtableThreshold to 2 > GB/CF even though i specified the large heap? > > -- > Thanks, > Mubarak Seyed.
