Page1 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: 2015
Gopal Vijayaraghavan
Page2 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC – Optimized Row-Columnar File
Columnar Storage+
Row-groups & Fixed splits
Protobuf Metadata Storage+
+
Type-safe Vectorization+
Hive ACID transactions+
Single SerDe for Format+
Page3 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Need for Speed: The Stinger Initiative
Stinger: An Open Roadmap to improve Apache Hive’s performance 100x.
Launched: February 2013; Delivered: April 2014.
Delivered in 100% Apache Open Source.
SQL Engine
Vectorized
SQL Engine
Columnar
Storage
ORC
= 100X+ +
Distributed
Execution
Apache Tez
Page4 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC at Facebook
Saved more than 1,400
servers worth of storage.
Compressioni
Compression ratio
increased from 5x to 8x
globally.
Compressioni
[1]
Page5 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC at Spotify
16x less HDFS read when
using ORC versus Avro.(5)
IOi
32x less CPU when using
ORC versus Avro.(5)
CPUi
[2]
Page 6 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: Today
What is Optimized about ORC?
Page7 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC – Optimized Row-Columnar File
Columnar Storage+
Row-groups & Stripe splits
Protobuf Metadata Storage+
+
Type-safe Vectorization+
Hive ACID transactions+
Single SerDe for Format+
Page8 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Columnar Storage
Storage Performance
● Compress each column differently
● Detect & compress common sub-sequences
● Auto-increment ids
● String Enums
● Large Integers (uid scale)
● Unique strings (UUIDS)
Read Performance
● Column projection
● Columnar deserializers
● Data locality
Write Throughput
● Stats auto-gather
Page9 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Row-groups & Stripe splits
Split Parallelism
● Effective parallelism
● No seeks to find boundaries
● No splits with zero data
● Decompress fixed chunks
Stripes
● Single unsplittable chunk
● Will reside in 1 HDFS block entirely
● Is self-contained for all read ops
Page10 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
A Single SerDe for all ORC Files
A Single Writer
● No mismatch of serialization
● Forward compatibility
Readers
● Multiple reader implementations
● Allows for vector readers
● And row-mode readers
● Similar loop – good JIT hit-rate
Page11 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Protobuf Metadata Storage
Standardized Metadata
● Readers are easier to write
● Metadata readers are auto-generated
Metadata Forward Compatibility
● Protobuf Optional fields
Statistics Storage in Metadata
● Standard serialization for stats
● Allows for PPD into the IO layer
Page12 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Type-safe Vectorization
Schema on Write
● Write ORC Structs with types
● SerDe & Inputformat
Read Performance
● Data is read with few copies
● Primitive types are fast
● Primitives are also unboxed
● Predicates are typed too
Page 13 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: ETL Improvements
Always more new data
Page14 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC (Zlib): Compress Differently
674
389
433
ORC (old zlib) ORC SNAPPY ORC (new zlib)
ETL for TPC-H LineItem (scale 1 Tb)
Time Taken
Different Zlib algorithms for encoding
● Z_FILTERED
● Z_DEFAULT
● Z_BEST_SPEED
● Z_DEFAULT_COMPRESSION
In detail
● Compress IS_NULL bitsets lightly
● Compress Integers differently from Doubles
● Compress string dictionaries differently
● Allow for user choice
Page15 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC (Zlib): Compress Differently
Different Zlib algorithms for encoding
● Z_FILTERED
● Z_DEFAULT
● Z_BEST_SPEED
● Z_DEFAULT_COMPRESSION
In detail
● Compress IS_NULL bitsets lightly
● Compress Integers differently from Doubles
● Compress string dictionaries differently
● Allow for user choice
178.5
225.1
172.2
ORC (old zlib) ORC SNAPPY ORC (new zlib)
Data Sizes for TPC-H Lineitem (Scale 1 Tb)
Size on Disk
Page16 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Using JDK8 SIMD: Integer Writers
Integer encodings
● Base + Delta
● Run-length
● Direct
Trade-off for Size/Speed
● Use fixed bit-width loops
● Snap to nearest bit-width
0
200
400
600
800
1000
1200
1400
1600
1800
2000
1 2 4 8 16 24 32 40 48 56 64
MeanTime(ms)
Bit Width
ORC Write Integer Performance
(smaller better)
hive 0.13 bitpacking
hive 1.0 bitpacking (new)
Page17 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Double Writers
273.331
247.634
231.741
0
50
100
150
200
250
300
old buffered + BE buffered + LE
MeanTime(ms)
Double Write Modes
ORC Write Double Performance
(smaller is better)
Double Writers
● JVM is big-endian
● X86 is little-endian
● Special handling of NaN
Page18 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: Scale compression buffers
269.4
263.3
258.5 258.4 258.4 258.4
184.8 183.5 182.2 180.1 178.3 177.4
140
160
180
200
220
240
260
280
300
320
8 16 32 64 128 256
SizeinMB
Compression Buffer Size in KB
File Size
ZLIB
SNAPPY
Large Columns vs More Columns
● Adjust when >1000 columns
Trade offs
● Compression
● Low memory use
More additions
● Dynamically partitioned insert
Page19 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: Streaming Ingest + ACID
Broken pattern: Partitions for Atomicity-
- Isolation & Consistency on retries+
Transactions are pluggable (txn.manager)+
Cache/Replication friendly (base + deltas)+
Page 20 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: LLAP and Sub-second
ORC – Pushing for Sub-second
Page21 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: Row Indexes
Min-Max pruning
● Evaluate on statistics
Bloom filters
● Better String filters
● Filter a random distribution
LLAP Future
● Row-level vector SARGs
5999989709
540,000
10,000
No Indexes Min-Max Indexes Bloomfilter Indexes
from tpch_1000.lineitem where l_orderkey = 1212000001;
(log scale)
Rows Read
Page22 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: Row Indexes
Min-Max pruning
● Evaluate on Statistics
Bloom filters
● Better String filters
● Filter a random distribution
LLAP Future
● Row-level vector SARGs
74
4.5 1.34
No Indexes Min-Max Indexes Bloomfilter Indexes
* from tpch_1000.lineitem where l_orderkey=1212000001;
(smaller better)
Time Taken (seconds)
Page23 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: JDK8 SIMD Readers
Integer encodings
● Base + Delta
● Run-length
● Direct
Trade-off for Size/Speed
● Use fixed bit-width loops
● Snap to nearest bit-width
0
200
400
600
800
1000
1200
1400
1600
1800
1 2 4 8 16 24 32 40 48 56 64
MeanTime(ms)
Bit Width
ORC Read Integer Performance
hive 0.13 unpacking
hive-1.0 unpacking (new)
Page24 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: Vectorization + SIMD
Advantage of a Single SerDe
● Primitive Types
Allocation free tight inner loops
● JDK8 has auto-vectorization
Vectorized Early Filter
● Vectors can be filtered early in ORC
● StringDictionary can be used to binary-search
Vectorized SIMD Join
● Performance for single key joins
0x00007f13d2e6afb0: vmovdqu 0x10(%rsi,%rax,8),%ymm2
0x00007f13d2e6afb6: vaddpd %ymm1,%ymm2,%ymm2
0x00007f13d2e6afba: movslq %eax,%r10
0x00007f13d2e6afbd: vmovdqu 0x30(%rsi,%r10,8),%ymm3
;*daload vector.expressions.gen.DoubleColAddDoubleColumn::evaluate
(line 94)
0x00007f13d2e6afc4: vmovdqu %ymm2,0x10(%rdx,%rax,8)
0x00007f13d2e6afca: vaddpd %ymm1,%ymm3,%ymm2
0x00007f13d2e6afce: vmovdqu %ymm2,0x30(%rdx,%r10,8)
;*dastore vector.expressions.gen.DoubleColAddDoubleColumn::evaluate
(line 94)
Page25 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: Split Strategies + Tez Grouping
Amdahl’s Law
● As fast as the slowest task
● Slice work thinly, but not too thin
Split-generation vs Execution time
● ETL
● BI
● Hybrid
Split-grouping & estimation
● ColumnarSplit size
● Group by estimate, not file size
● Bucket pruning
Slow split
Page26 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: LLAP
- JIT Performance for short queries+
Row-group level caching+
Asynchronous IO Elevator+
+ Multi-threaded Column Vector processing+
Page27 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
ORC: LLAP (+ SIMD + Split Strategies + Row Indexes)
Page28 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Questions?
?
Interested? Stop by the Hortonworks booth to learn more
Page29 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Endnotes
(1) https://code.facebook.com/posts/229861827208629/scaling-the-facebook-data-warehouse-to-300-pb/
(2) http://www.slideshare.net/AdamKawa/a-perfect-hive-query-for-a-perfect-meeting-hadoop-summit-2014

ORC 2015

  • 1.
    Page1 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: 2015 Gopal Vijayaraghavan
  • 2.
    Page2 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC – Optimized Row-Columnar File Columnar Storage+ Row-groups & Fixed splits Protobuf Metadata Storage+ + Type-safe Vectorization+ Hive ACID transactions+ Single SerDe for Format+
  • 3.
    Page3 © HortonworksInc. 2011 – 2015. All Rights Reserved Need for Speed: The Stinger Initiative Stinger: An Open Roadmap to improve Apache Hive’s performance 100x. Launched: February 2013; Delivered: April 2014. Delivered in 100% Apache Open Source. SQL Engine Vectorized SQL Engine Columnar Storage ORC = 100X+ + Distributed Execution Apache Tez
  • 4.
    Page4 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC at Facebook Saved more than 1,400 servers worth of storage. Compressioni Compression ratio increased from 5x to 8x globally. Compressioni [1]
  • 5.
    Page5 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC at Spotify 16x less HDFS read when using ORC versus Avro.(5) IOi 32x less CPU when using ORC versus Avro.(5) CPUi [2]
  • 6.
    Page 6 ©Hortonworks Inc. 2011 – 2015. All Rights Reserved ORC: Today What is Optimized about ORC?
  • 7.
    Page7 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC – Optimized Row-Columnar File Columnar Storage+ Row-groups & Stripe splits Protobuf Metadata Storage+ + Type-safe Vectorization+ Hive ACID transactions+ Single SerDe for Format+
  • 8.
    Page8 © HortonworksInc. 2011 – 2015. All Rights Reserved Columnar Storage Storage Performance ● Compress each column differently ● Detect & compress common sub-sequences ● Auto-increment ids ● String Enums ● Large Integers (uid scale) ● Unique strings (UUIDS) Read Performance ● Column projection ● Columnar deserializers ● Data locality Write Throughput ● Stats auto-gather
  • 9.
    Page9 © HortonworksInc. 2011 – 2015. All Rights Reserved Row-groups & Stripe splits Split Parallelism ● Effective parallelism ● No seeks to find boundaries ● No splits with zero data ● Decompress fixed chunks Stripes ● Single unsplittable chunk ● Will reside in 1 HDFS block entirely ● Is self-contained for all read ops
  • 10.
    Page10 © HortonworksInc. 2011 – 2015. All Rights Reserved A Single SerDe for all ORC Files A Single Writer ● No mismatch of serialization ● Forward compatibility Readers ● Multiple reader implementations ● Allows for vector readers ● And row-mode readers ● Similar loop – good JIT hit-rate
  • 11.
    Page11 © HortonworksInc. 2011 – 2015. All Rights Reserved Protobuf Metadata Storage Standardized Metadata ● Readers are easier to write ● Metadata readers are auto-generated Metadata Forward Compatibility ● Protobuf Optional fields Statistics Storage in Metadata ● Standard serialization for stats ● Allows for PPD into the IO layer
  • 12.
    Page12 © HortonworksInc. 2011 – 2015. All Rights Reserved Type-safe Vectorization Schema on Write ● Write ORC Structs with types ● SerDe & Inputformat Read Performance ● Data is read with few copies ● Primitive types are fast ● Primitives are also unboxed ● Predicates are typed too
  • 13.
    Page 13 ©Hortonworks Inc. 2011 – 2015. All Rights Reserved ORC: ETL Improvements Always more new data
  • 14.
    Page14 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC (Zlib): Compress Differently 674 389 433 ORC (old zlib) ORC SNAPPY ORC (new zlib) ETL for TPC-H LineItem (scale 1 Tb) Time Taken Different Zlib algorithms for encoding ● Z_FILTERED ● Z_DEFAULT ● Z_BEST_SPEED ● Z_DEFAULT_COMPRESSION In detail ● Compress IS_NULL bitsets lightly ● Compress Integers differently from Doubles ● Compress string dictionaries differently ● Allow for user choice
  • 15.
    Page15 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC (Zlib): Compress Differently Different Zlib algorithms for encoding ● Z_FILTERED ● Z_DEFAULT ● Z_BEST_SPEED ● Z_DEFAULT_COMPRESSION In detail ● Compress IS_NULL bitsets lightly ● Compress Integers differently from Doubles ● Compress string dictionaries differently ● Allow for user choice 178.5 225.1 172.2 ORC (old zlib) ORC SNAPPY ORC (new zlib) Data Sizes for TPC-H Lineitem (Scale 1 Tb) Size on Disk
  • 16.
    Page16 © HortonworksInc. 2011 – 2015. All Rights Reserved Using JDK8 SIMD: Integer Writers Integer encodings ● Base + Delta ● Run-length ● Direct Trade-off for Size/Speed ● Use fixed bit-width loops ● Snap to nearest bit-width 0 200 400 600 800 1000 1200 1400 1600 1800 2000 1 2 4 8 16 24 32 40 48 56 64 MeanTime(ms) Bit Width ORC Write Integer Performance (smaller better) hive 0.13 bitpacking hive 1.0 bitpacking (new)
  • 17.
    Page17 © HortonworksInc. 2011 – 2015. All Rights Reserved Double Writers 273.331 247.634 231.741 0 50 100 150 200 250 300 old buffered + BE buffered + LE MeanTime(ms) Double Write Modes ORC Write Double Performance (smaller is better) Double Writers ● JVM is big-endian ● X86 is little-endian ● Special handling of NaN
  • 18.
    Page18 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: Scale compression buffers 269.4 263.3 258.5 258.4 258.4 258.4 184.8 183.5 182.2 180.1 178.3 177.4 140 160 180 200 220 240 260 280 300 320 8 16 32 64 128 256 SizeinMB Compression Buffer Size in KB File Size ZLIB SNAPPY Large Columns vs More Columns ● Adjust when >1000 columns Trade offs ● Compression ● Low memory use More additions ● Dynamically partitioned insert
  • 19.
    Page19 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: Streaming Ingest + ACID Broken pattern: Partitions for Atomicity- - Isolation & Consistency on retries+ Transactions are pluggable (txn.manager)+ Cache/Replication friendly (base + deltas)+
  • 20.
    Page 20 ©Hortonworks Inc. 2011 – 2015. All Rights Reserved ORC: LLAP and Sub-second ORC – Pushing for Sub-second
  • 21.
    Page21 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: Row Indexes Min-Max pruning ● Evaluate on statistics Bloom filters ● Better String filters ● Filter a random distribution LLAP Future ● Row-level vector SARGs 5999989709 540,000 10,000 No Indexes Min-Max Indexes Bloomfilter Indexes from tpch_1000.lineitem where l_orderkey = 1212000001; (log scale) Rows Read
  • 22.
    Page22 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: Row Indexes Min-Max pruning ● Evaluate on Statistics Bloom filters ● Better String filters ● Filter a random distribution LLAP Future ● Row-level vector SARGs 74 4.5 1.34 No Indexes Min-Max Indexes Bloomfilter Indexes * from tpch_1000.lineitem where l_orderkey=1212000001; (smaller better) Time Taken (seconds)
  • 23.
    Page23 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: JDK8 SIMD Readers Integer encodings ● Base + Delta ● Run-length ● Direct Trade-off for Size/Speed ● Use fixed bit-width loops ● Snap to nearest bit-width 0 200 400 600 800 1000 1200 1400 1600 1800 1 2 4 8 16 24 32 40 48 56 64 MeanTime(ms) Bit Width ORC Read Integer Performance hive 0.13 unpacking hive-1.0 unpacking (new)
  • 24.
    Page24 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: Vectorization + SIMD Advantage of a Single SerDe ● Primitive Types Allocation free tight inner loops ● JDK8 has auto-vectorization Vectorized Early Filter ● Vectors can be filtered early in ORC ● StringDictionary can be used to binary-search Vectorized SIMD Join ● Performance for single key joins 0x00007f13d2e6afb0: vmovdqu 0x10(%rsi,%rax,8),%ymm2 0x00007f13d2e6afb6: vaddpd %ymm1,%ymm2,%ymm2 0x00007f13d2e6afba: movslq %eax,%r10 0x00007f13d2e6afbd: vmovdqu 0x30(%rsi,%r10,8),%ymm3 ;*daload vector.expressions.gen.DoubleColAddDoubleColumn::evaluate (line 94) 0x00007f13d2e6afc4: vmovdqu %ymm2,0x10(%rdx,%rax,8) 0x00007f13d2e6afca: vaddpd %ymm1,%ymm3,%ymm2 0x00007f13d2e6afce: vmovdqu %ymm2,0x30(%rdx,%r10,8) ;*dastore vector.expressions.gen.DoubleColAddDoubleColumn::evaluate (line 94)
  • 25.
    Page25 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: Split Strategies + Tez Grouping Amdahl’s Law ● As fast as the slowest task ● Slice work thinly, but not too thin Split-generation vs Execution time ● ETL ● BI ● Hybrid Split-grouping & estimation ● ColumnarSplit size ● Group by estimate, not file size ● Bucket pruning Slow split
  • 26.
    Page26 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: LLAP - JIT Performance for short queries+ Row-group level caching+ Asynchronous IO Elevator+ + Multi-threaded Column Vector processing+
  • 27.
    Page27 © HortonworksInc. 2011 – 2015. All Rights Reserved ORC: LLAP (+ SIMD + Split Strategies + Row Indexes)
  • 28.
    Page28 © HortonworksInc. 2011 – 2015. All Rights Reserved Questions? ? Interested? Stop by the Hortonworks booth to learn more
  • 29.
    Page29 © HortonworksInc. 2011 – 2015. All Rights Reserved Endnotes (1) https://code.facebook.com/posts/229861827208629/scaling-the-facebook-data-warehouse-to-300-pb/ (2) http://www.slideshare.net/AdamKawa/a-perfect-hive-query-for-a-perfect-meeting-hadoop-summit-2014