Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
1.13.0
-
None
Description
Current Spark use Parquet-mr as parquet reader/writer library, but the built-in bit-packing en/decode is not efficient enough.
Our optimization for Parquet bit-packing en/decode with jdk.incubator.vector in Open JDK18 brings prominent performance improvement.
Due to Vector API is added to OpenJDK since 16, So this optimization request JDK16 or higher.
Below are our test results
Functional test is based on open-source parquet-mr Bit-pack decoding function: public final void unpack8Values(final byte[] in, final int inPos, final int[] out, final int outPos) __
compared with our implementation with vector API public final void unpack8Values_vec(final byte[] in, final int inPos, final int[] out, final int outPos)
We tested 10 pairs (open source parquet bit unpacking vs ours optimized vectorized SIMD implementation) decode function with bit width={1,2,3,4,5,6,7,8,9,10}, below are test results:
We integrated our bit-packing decode implementation into parquet-mr, tested the parquet batch reader ability from Spark VectorizedParquetRecordReader which get parquet column data by the batch way. We construct parquet file with different row count and column count, the column data type is Int32, the maximum int value is 127 which satisfies bit pack encode with bit width=7, the count of the row is from 10k to 100 million and the count of the column is from 1 to 4.
Attachments
Attachments
Issue Links
- is a parent of
-
PARQUET-2375 Extend vectorized bit unpacking benchmark for various bit sizes.
- Resolved
1.
|
Parquet java vector decode optimization for Big Endian | Open | Unassigned | |
2.
|
Parquet java vector decode optimization Long for Big Endian | Open | Unassigned | |
3.
|
Parquet java vector decode optimization Long for Little Endian | Open | Unassigned |