[
https://issues.apache.org/jira/browse/SPARK-11583?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14996147#comment-14996147
]
Kent Yao commented on SPARK-11583:
----------------------------------
[~srowen] MapStatus stores BlockManagerId where it locates and Array[Long]
which are the sizes of shuffle results, One of its implementation
HighlyCompressedMapStatus compresses the Array[long] to a *long* ---avgSize and
a BitSet[Array.size] marking empty blocks. For large tasks such as 200k, the
size of bitset[200k] is 128KB, when 200k mapstatuses be passed to the diver, it
will be 200k * 128KB ≈ 4.66GB, each executor will fetch them later. What if
only few blocks contain result , or only few blocks are empty (maybe only 10),
we use a container stores the _reduceId: Int_ , size of such a mapstatus may be
10*32bit, which means the diver may only receive 61MB data.
Am I right?
> Make MapStatus use less memory uage
> -----------------------------------
>
> Key: SPARK-11583
> URL: https://issues.apache.org/jira/browse/SPARK-11583
> Project: Spark
> Issue Type: Improvement
> Reporter: Kent Yao
>
> In the resolved issue https://issues.apache.org/jira/browse/SPARK-11271, as I
> said, using BitSet can save ≈20% memory usage compared to RoaringBitMap.
> For a spark job contains quite a lot of tasks, 20% seems a drop in the ocean.
> Essentially, BitSet uses long[]. For example a BitSet[200k] = long[3125].
> So if we use a HashSet[Int] to store reduceId (when non-empty blocks are
> dense,use reduceId of empty blocks; when sparse, use non-empty ones).
> For dense cases: if HashSet[Int](numNonEmptyBlocks).size <
> BitSet[totalBlockNum], I use MapStatusTrackingNoEmptyBlocks
> For sparse cases: if HashSet[Int](numEmptyBlocks).size <
> BitSet[totalBlockNum], I use MapStatusTrackingEmptyBlocks
> sparse case, 299/300 are empty
> sc.makeRDD(1 to 30000, 3000).groupBy(x=>x).top(5)
> dense case, no block is empty
> sc.makeRDD(1 to 9000000, 3000).groupBy(x=>x).top(5)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
