KQR L PO stands for *K*ernel *Q*uery layer *R*ow cache *L*arge *P*arent *O*bject. These are dictionary cache objects. So, you either have a row cache object leak (less likely) or just a lot of objects to keep in row cache (row cache = dictionary cache). For example, if you have a lots of tables with lots of columns and all of these columns have histograms, then you'll have lots of histogram data cached in dictionary cache's *dc_histogram_defs* section. So, you should run something like this, to see what kind of rowcache objects use all this memory: select * from v$rowcache order by usage; Once you find the top consumer, you can use AWR's DBA_HIST_ROWCACHE_SUMMARY to see how these values have changed over time... -- Tanel Poder New virtual conference and online seminars! http://tech.e2sn.com/virtual-conferences http://tech.e2sn.com/oracle-training-seminars On Mon, Oct 18, 2010 at 8:49 PM, Grzegorz Goryszewski <grzegorzof@xxxxxxxxxx > wrote: > Google and MOS are quite mysterious about that . > KQR L PO is huge allocation in my RAC shared pool , and I dont got idea > what could cause that . > Regards. > Greg >