Snippets Groups Projects

[SPARK-14724] Use radix sort for shuffles and sort operator when possible

Eric Liang authored 9 years ago

## What changes were proposed in this pull request?

Spark currently uses TimSort for all in-memory sorts, including sorts done for shuffle. One low-hanging fruit is to use radix sort when possible (e.g. sorting by integer keys). This PR adds a radix sort implementation to the unsafe sort package and switches shuffles and sorts to use it when possible.

The current implementation does not have special support for null values, so we cannot radix-sort `LongType`. I will address this in a follow-up PR.

## How was this patch tested?

Unit tests, enabling radix sort on existing tests. Microbenchmark results:

```
Running benchmark: radix sort 25000000
Java HotSpot(TM) 64-Bit Server VM 1.8.0_66-b17 on Linux 3.13.0-44-generic
Intel(R) Core(TM) i7-4600U CPU  2.10GHz

radix sort 25000000:                Best/Avg Time(ms)    Rate(M/s)   Per Row(ns)   Relative
-------------------------------------------------------------------------------------------
reference TimSort key prefix array     15546 / 15859          1.6         621.9       1.0X
reference Arrays.sort                    2416 / 2446         10.3          96.6       6.4X
radix sort one byte                       133 /  137        188.4           5.3     117.2X
radix sort two bytes                      255 /  258         98.2          10.2      61.1X
radix sort eight bytes                    991 /  997         25.2          39.6      15.7X
radix sort key prefix array              1540 / 1563         16.2          61.6      10.1X
```

I also ran a mix of the supported TPCDS queries and compared TimSort vs RadixSort metrics. The overall benchmark ran ~10% faster with radix sort on. In the breakdown below, the radix-enabled sort phases averaged about 20x faster than TimSort, however sorting is only a small fraction of the overall runtime. About half of the TPCDS queries were able to take advantage of radix sort.

```
TPCDS on master: 2499s real time, 8185s executor
    - 1171s in TimSort, avg 267 MB/s
(note the /s accounting is weird here since dataSize counts the record sizes too)

TPCDS with radix enabled: 2294s real time, 7391s executor
    - 596s in TimSort, avg 254 MB/s
    - 26s in radix sort, avg 4.2 GB/s
```

cc davies rxin

Author: Eric Liang <ekl@databricks.com>

Closes #12490 from ericl/sort-benchmark.

e2b5647a

e2b5647a 9 years ago

Name	Last commit	Last update
.github
R
assembly
bin
build
common
conf
core
data
dev
docs
examples
external
graphx
launcher
licenses
mllib-local
mllib
project
python
repl
sbin
sql
streaming
tools
yarn
.gitattributes
.gitignore
CONTRIBUTING.md
LICENSE
NOTICE
README.md
pom.xml
scalastyle-config.xml

Apache Spark

Spark is a fast and general cluster computing system for Big Data. It provides high-level APIs in Scala, Java, Python, and R, and an optimized engine that supports general computation graphs for data analysis. It also supports a rich set of higher-level tools including Spark SQL for SQL and DataFrames, MLlib for machine learning, GraphX for graph processing, and Spark Streaming for stream processing.

http://spark.apache.org/

Online Documentation

You can find the latest Spark documentation, including a programming guide, on the project web page and project wiki. This README file only contains basic setup instructions.

Building Spark

Spark is built using Apache Maven. To build Spark and its example programs, run:

build/mvn -DskipTests clean package

(You do not need to do this if you downloaded a pre-built package.) More detailed documentation is available from the project site, at "Building Spark". For developing Spark using an IDE, see Eclipse and IntelliJ.

Interactive Scala Shell

The easiest way to start using Spark is through the Scala shell:

./bin/spark-shell

Try the following command, which should return 1000:

scala> sc.parallelize(1 to 1000).count()

Interactive Python Shell

Alternatively, if you prefer Python, you can use the Python shell:

./bin/pyspark

And run the following command, which should also return 1000:

>>> sc.parallelize(range(1000)).count()

Example Programs

Spark also comes with several sample programs in the examples directory. To run one of them, use ./bin/run-example <class> [params]. For example:

./bin/run-example SparkPi

will run the Pi example locally.

You can set the MASTER environment variable when running examples to submit examples to a cluster. This can be a mesos:// or spark:// URL, "yarn" to run on YARN, and "local" to run locally with one thread, or "local[N]" to run locally with N threads. You can also use an abbreviated class name if the class is in the examples package. For instance:

MASTER=spark://host:7077 ./bin/run-example SparkPi

Many of the example programs print usage help if no params are given.

Running Tests

Testing first requires building Spark. Once Spark is built, tests can be run using:

./dev/run-tests

Please see the guidance on how to run tests for a module, or individual tests.

A Note About Hadoop Versions

Spark uses the Hadoop core library to talk to HDFS and other Hadoop-supported storage systems. Because the protocols have changed in different versions of Hadoop, you must build Spark against the same version that your cluster runs.

Please refer to the build documentation at "Specifying the Hadoop Version" for detailed guidance on building for a particular distribution of Hadoop, including building for particular Hive and Hive Thriftserver distributions.

Configuration

Please refer to the Configuration Guide in the online documentation for an overview on how to configure Spark.