Thanks Mike, I have 25 millions docs indexed, faceted on simple fields
(cardinality: 5 for country field and 10000 for host field)
8192Mb, JRockit R27 (Java 6)
Unpredictable OOMs...
I set HashDocSet/max to 30,000, don't see any performance degradation
yet (the same response times for faceted id:[* TO *] and some queries).
Cache is small (trying to avoid OOM):
<filterCache
class="solr.LRUCache"
size="1000"
initialSize="100"
autowarmCount="0"/>
<queryResultCache
class="solr.LRUCache"
size="1000"
initialSize="100"
autowarmCount="0"/>
<documentCache
class="solr.LRUCache"
size="1000"
initialSize="100"
autowarmCount="0"/>
Quoting Mike Klaas <[EMAIL PROTECTED]>:
On 17-Jul-08, at 10:28 AM, Fuad Efendi wrote:
<!-- This entry enables an int hash representation for filters (DocSets)
when the number of items in the set is less than maxSize.
For smaller
sets, this representation is more memory efficient, more efficient to
iterate over, and faster to take intersections. -->
<HashDocSet maxSize="3000" loadFactor="0.75"/>
Change it to higher value, for instance, 30000. OpenBitSet is
created for larger values and requires a lot of memory...
Careful--hash sets of that size can be quite slow. It does make sense
to bump up the value to 6000 or so for large (multi-million) indices.
-Mike
