spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Anton Okolnychyi (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (SPARK-26205) Optimize InSet expression for bytes, shorts, ints, dates
Date Thu, 28 Feb 2019 14:11:00 GMT

     [ https://issues.apache.org/jira/browse/SPARK-26205?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Anton Okolnychyi updated SPARK-26205:
-------------------------------------
    Description: 
{{In}} expressions are compiled into a sequence of if-else statements, which results in O\(n\)
time complexity. {{InSet}} is an optimized version of {{In}}, which is supposed to improve
the performance if the number of elements is big enough. However, {{InSet}} actually degrades
the performance in many cases due to various reasons (benchmarks were created in SPARK-26203
and solutions to the boxing problem are discussed in SPARK-26204).

The main idea of this JIRA is to use Java {{switch}} statements to significantly improve the
performance of {{InSet}} expressions for bytes, shorts, ints, dates. All {{switch}} statements
are compiled into {{tableswitch}} and {{lookupswitch}} bytecode instructions. We will have
O\(1\) time complexity if our case values are compact and {{tableswitch}} can be used. Otherwise,
{{lookupswitch}} will give us O\(log n\). Our local benchmarks show that this logic is more
than two times faster even on 500+ elements than using primitive collections in {{InSet}}
expressions. As Spark is using Scala {{HashSet}} right now, the performance gain will be is
even bigger.

See [here|https://docs.oracle.com/javase/specs/jvms/se7/html/jvms-3.html#jvms-3.10] and [here|https://stackoverflow.com/questions/10287700/difference-between-jvms-lookupswitch-and-tableswitch]
for more information.

  was:
Currently, {{In}} expressions are compiled into a sequence of if-else statements, which results
in O\(n\) time complexity. {{InSet}} is an optimized version of {{In}}, which is supposed
to improve the performance if the number of elements is big enough. However, {{InSet}} actually
degrades the performance in many cases due to various reasons (benchmarks will be available
in SPARK-26203 and solutions are discussed in SPARK-26204).

The main idea of this JIRA is to make use of {{tableswitch}} and {{lookupswitch}} bytecode instructions.
In short, we can improve our time complexity from O\(n\) to O\(1\) or at least O\(log n\)
by using Java {{switch}} statements. We will have O\(1\) time complexity if our case values
are compact and {{tableswitch}} can be used. Otherwise, {{lookupswitch}} will give us O\(log
n\). 

An important benefit of the proposed approach is that we do not have to pay an extra cost
for autoboxing as in case of {{InSet}}. As a consequence, we can substantially outperform
{{InSet}} even on 250+ elements.

See [here|https://docs.oracle.com/javase/specs/jvms/se7/html/jvms-3.html#jvms-3.10] and [here|https://stackoverflow.com/questions/10287700/difference-between-jvms-lookupswitch-and-tableswitch]
for more information.


> Optimize InSet expression for bytes, shorts, ints, dates
> --------------------------------------------------------
>
>                 Key: SPARK-26205
>                 URL: https://issues.apache.org/jira/browse/SPARK-26205
>             Project: Spark
>          Issue Type: Improvement
>          Components: SQL
>    Affects Versions: 3.0.0
>            Reporter: Anton Okolnychyi
>            Priority: Major
>
> {{In}} expressions are compiled into a sequence of if-else statements, which results
in O\(n\) time complexity. {{InSet}} is an optimized version of {{In}}, which is supposed
to improve the performance if the number of elements is big enough. However, {{InSet}} actually
degrades the performance in many cases due to various reasons (benchmarks were created in
SPARK-26203 and solutions to the boxing problem are discussed in SPARK-26204).
> The main idea of this JIRA is to use Java {{switch}} statements to significantly improve
the performance of {{InSet}} expressions for bytes, shorts, ints, dates. All {{switch}} statements
are compiled into {{tableswitch}} and {{lookupswitch}} bytecode instructions. We will have
O\(1\) time complexity if our case values are compact and {{tableswitch}} can be used. Otherwise,
{{lookupswitch}} will give us O\(log n\). Our local benchmarks show that this logic is more
than two times faster even on 500+ elements than using primitive collections in {{InSet}}
expressions. As Spark is using Scala {{HashSet}} right now, the performance gain will be is
even bigger.
> See [here|https://docs.oracle.com/javase/specs/jvms/se7/html/jvms-3.html#jvms-3.10] and
[here|https://stackoverflow.com/questions/10287700/difference-between-jvms-lookupswitch-and-tableswitch]
for more information.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org


Mime
View raw message