[infinispan-dev] Do we need bulk eviction notification?
Mircea Markus
mircea.markus at jboss.com
Thu Oct 14 16:18:49 EDT 2010
On 14 Oct 2010, at 14:53, Vladimir Blagojevic wrote:
> I think this is a great idea Trustin! Do you want to do this through JIRA?
+1
> On 2010-10-14, at 7:20 AM, 이희승 (Trustin Lee) wrote:
>
>> Current BoundedConcurrentHashMap notifies EvictionListener for every
>> evicted entry. However, as you see from
>> BoundedConcurrentHashMap.Segment.attemptEviction(..), eviction often
>> happens in bulk.
>>
>> If these evicted entries are supposed to be passivated, it is often
>> beneficial to begin a transaction because it often takes much less time
>> to call commit() less often. If N entries are evicted:
>>
>> without bulk eviction:
>>
>> store.store(e1); -- each store implied a commit()
>> store.store(..);
>> store.store(eN);
>>
>> with bulk eviction:
>>
>> store.begin(mods(e1, ..., eN), tx, true);
>>
>> so, what do you think about changing the EvictionListener interface like
>> this:
>>
>> interface EvictionListener<K, V> {
>> void preEvict(K key);
>> void postEvict(K key, V value);
>>
>> void preBulkEvict(Set<K> keys);
>> void postBulkEvict(Map<K, V> entries);
>> }
>>
>> abstract class AbstractEvictionListener<K, V>
>> implements EvictionListener {
>> void preBulkEvict(Set<K> keys) {
>> for (K k: keys) { preEvict(k); }
>> }
>> void postBulkEvict(Set<K> keys) {
>> ...
>> }
>> }
>>
>> I might be missing something though. Please let me know if there is a
>> way to achieve similar improvement without this modification.
>>
>> --
>> Trustin Lee - http://gleamynode.net/
>>
>>
>> _______________________________________________
>> infinispan-dev mailing list
>> infinispan-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>
> --
> Vladimir Blagojevic
> JBoss Clustering Team
> JBoss, by Red Hat
>
>
>
>
>
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev
More information about the infinispan-dev
mailing list