[
https://issues.apache.org/jira/browse/PHOENIX-2126?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14699630#comment-14699630
]
Ankit Singhal commented on PHOENIX-2126:
----------------------------------------
Hi [~rajeshbabu],
PFA, the updated patch for PHOENIX-2126_v2.0.patch
All the below review comments are incorporated now.
[x] Need to use dev/PhoenixCodeTemplate.xml template for formatting a code.
[x] Make number of threads configurable depending upon the number of
iterators/guideposts and take number of CPU in a account for parallelization.
[x] Try to use the same thread pool used for querying.
[x] Add a unit case for reverse sort also.
[x] Need to create a patch from master as earlier patch was created from 4.2
branch
[x] Check for OOM errors when huge data set returned from servers.
For avoiding OOM errors , the queue is changed to MappedByteBufferTupleQueue
but it could have performance problems so need benchmark its impact.
One question :- is it possible move out MappedByteBufferTupleQueue class from
SortMergeJoinPlan as it is currently private and can not be externally?
Regards,
Ankit Singhal
> Improving performance of merge sort by multi-threaded and minheap
> implementation
> --------------------------------------------------------------------------------
>
> Key: PHOENIX-2126
> URL: https://issues.apache.org/jira/browse/PHOENIX-2126
> Project: Phoenix
> Issue Type: Improvement
> Affects Versions: 4.1.0, 4.2.0
> Reporter: Ankit Singhal
> Attachments: PHOENIX-2126_v1.0.patch, PHOENIX-2126_v2.0.patch
>
>
> {code}
> CREATE TABLE IF NOT EXISTS test (
> dim1 INTEGER NOT NULL,
> A.B INTEGER,
> A.M DECIMAL,
> CONSTRAINT PK PRIMARY KEY
> (dim1))
> SALT_BUCKETS =256,DEFAULT_COLUMN_FAMILY='A';
> {code}
> *Query to benchmark:-*
> {code}
> select dim1,sum(b),sum(m) from test where Datemth>=201505 and Datemth<=201505
> and dim1 IS NOT NULL group by dim1 order by sum(m) desc nulls last limit 10;
> {code}
> *current scenario:-*
> *CASE 1: * consider the case when dim1 is high cardinality attribute (10K+)
> and table have salt bucket set to 256, we will get 256 iterators from above
> query at the client and MergeSortRowKeyResultIterator has to merge these 256
> iterators with single thread. So let's say each iterator has 10k tuples
> returned, then merge sort needs to merge 2.5M tuples which will be costly if
> it is done with single thread and the query spend most of its time on client
> *CASE 2: * consider the case when dim1 is high cardinality attribute (10K+)
> and table have salt bucket set to 1, we will get 1 iterator from above query
> at the client and MergeSortRowKeyResultIterator doesn't need to merge
> anything. Here, it is fine with single thread.
> *CASE 3: * consider the case when dim1 is low cardinality attribute (10-100)
> and table have salt bucket set to 256, we will get 256 iterator from above
> query at the client and MergeSortRowKeyResultIterator has to merge these 256
> iterators with single thread. here the single thread is also fine as he has
> to merge only 2560 tuples.
> *Solution for case1 problem is:-*
> Optimized the implementation of merging 'n'-sorted iterators(having 'm'
> tuples) by using "min heap" which optimizes the time complexity from
> O(n2m) to O(nmLogn) (as heapify takes (Logn) time).
> And, By using multiple-threads('t') to process group of iterators which
> further optimized the complexity to
> T(nm)=T(nm)/t+T(t)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
