[ 
https://issues.apache.org/jira/browse/IGNITE-14076?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Stanilovsky Evgeny updated IGNITE-14076:
----------------------------------------
        Fix Version/s:     (was: 2.11)
    Affects Version/s:     (was: 2.9.1)
                       2.10

> Quadratic putAll performance degradation in transactional cache
> ---------------------------------------------------------------
>
>                 Key: IGNITE-14076
>                 URL: https://issues.apache.org/jira/browse/IGNITE-14076
>             Project: Ignite
>          Issue Type: Improvement
>          Components: cache
>    Affects Versions: 2.10
>            Reporter: Pavel Tupitsyn
>            Assignee: Stanilovsky Evgeny
>            Priority: Critical
>          Time Spent: 0.5h
>  Remaining Estimate: 0h
>
> {{putAll}} execution time grows almost exponentially while the number of keys 
> grows linearly in the following test:
> {code:java}
> public class PutAllTxTest extends GridCommonAbstractTest {
>     @Test
>     public void testPutAll() throws Exception {
>         Ignition.start(getConfiguration("server1"));
>         Ignition.start(getConfiguration("server2"));
>         Ignite ignite = 
> Ignition.start(getConfiguration("client").setClientMode(true));
>         IgniteCache<Integer, Integer> cache = ignite.createCache(
>                 new CacheConfiguration<Integer, Integer>("c")
>                         .setAtomicityMode(CacheAtomicityMode.TRANSACTIONAL));
>         int count = 50000;
>         Map<Integer, Integer> data = new TreeMap<>();
>         for (int i = 0; i < count; i++)
>             data.put(i, i);
>         long begin = System.nanoTime();
>         cache.putAll(data);
>         long dur = System.nanoTime() - begin;
>         System.out.println(">>>>> " + dur / 1000000);
>     }
> }
> {code}
> ||Entries||Seconds||
> |1000|0.4|
> |5000|1.9|
> |10000|3.8|
> |20000|10.7|
> |30000|23.5|
> |40000|41|
> |50000|64|
> |60000|90|
> |100000|254|
> This does not reproduce with 1 server node, and does not reproduce on 
> {{ATOMIC}} caches with any number of nodes.
> *Observations:*
> - Not a GC issue (it barely runs)
> - Not a memory issue (heap is under 1GB)
> - GridDhtTxPrepareFuture#localDhtPendingVersions -> 
> GridCacheMapEntry.localCandidates is the bottleneck. For 1K keys, 
> localCandidates gets called 123K times, 2K keys - 484K times, etc.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to