Five cool ways the JVM can run Apache Spark faster

© 2015 IBM Corporation
Five cool ways the JVM can run Apache Spark faster
Tim Ellison
IBM Runtime Technology Center, Hursley, UK
tellison
@tpellison

What's so cool about IBM's Java implementation?
Reference
Java
Technology
OpenJDK,
etc
IBM
Java
IBM Runtime
Technology
Centre
Quality
Engineering
Performance
Security
Reliability
Scalability
Serviceability
Production
Requirements
Strategic Software
Hardware exploits
User Feedback
 A certified Java, optimized to run on wide
variety of hardware platforms
 Performance tuned for key software workloads
 World class service and support
http://www.ibm.com/developerworks/java/jdk/index.html
Java Platform APIs
AWT Swing Java2d
XML Crypto Serializer
Core Libraries
Virtual Machine
Intel / POWER / z Series / ARM
Spark Backend
Spark APIs
Scala

© 2015 IBM Corporation3
Five cool things you and I can do with the JVM to enhance the
performance of Apache Spark!
1) JIT compiler enhancements and writing JIT-friendly code
2) Improving the object serializer
3) Faster IO – networking and storage
4) Data access acceleration & auto SIMD
5) Offloading tasks to graphics co-processors (GPUs)

input/output
bound jobs

processing
bound jobs

JIT compiler enhancements, and writing JIT-friendly code

Help yourself. JNI calls are not free!
https://github.com/xerial/snappyjava/blob/develop/src/main/java/org/xerial/snappy/SnappyNative.cpp

Style: Using JNI has an impact...
 The cost of calling from Java code to natives and from natives to Java code is significantly
higher (maybe 5x longer) than a normal Java method call.
– The JIT can't in-line native methods.
– The JIT can't do data flow analysis into JNI calls
• e.g. it has to assume that all parameters always used.
– The JIT has to set up the call stack and parameters for C calling convention,
• i.e. maybe rearranging items on the stack.
 JNI can introduce additional data copying costs
– There's no guarantee that you will get a direct pointer to the array / string with
Get<type>ArrayElements(), even when using the GetPrimitiveArrayCritical
versions.
– The IBM JVM will always return a copy (to allow GC to continue).
 Tip:
– JNI natives are more expensive than plain Java calls.
– e.g. create an unsafe based Snappy-like package written in Java code so that JNI cost is
eliminated.

Style: Use JIT optimizations to reduce overhead of logging checks
 Tip: Check for the non-null value of a static field ref to instance of a logging class singleton
– e.g.
– Uses the JIT's speculative optimization to avoid the explicit test for logging being enabled;
instead it ...
1)Generates an internal JIT runtime assumption (e.g. InfoLogger.class is undefined),
2)NOPs the test for trace enablement
3)Uses a class initialization hook for the InfoLogger.class (already necessary for instantiating the class)
4)The JIT will regenerate the test code if the class event is fired
 Spark's logging calls are gated on the checks of a static boolean value
trait Logging
Spark

Style: Judicious use of polymorphism
 Spark has a number of highly polymorphic interface call sites and high fan-in (several calling contexts
invoking the same callee method) in map, reduce, filter, flatMap, ...
– e.g. ExternalSorter.insertAll is very hot (drains an iterator using hasNext/next calls)
 Pattern #1:
– InterruptibleIterator → Scala's mapIterator → Scala's filterIterator → …
 Pattern #2:
– InterruptibleIterator → Scala's filterIterator → Scala's mapIterator → …
 The JIT can only choose one pattern to in-line!
– Makes JIT devirtualization and speculation more risky; using profiling information from a different
context could lead to incorrect devirtualization.
– More conservative speculation, or good phase change detection and recovery are needed in the JIT
compiler to avoid getting it wrong.
 Lambdas and functions as arguments, by definition, introduce different code flow targets
– Passing in widely implemented interfaces produce many different bytecode sequences
– When we in-line we have to put runtime checks ahead of in-lined method bodies to make sure we are
going to run the right method!
– Often specialized classes are used only in a very limited number of places, but the majority of the code
does not use these classes and pays a heavy penalty
– e.g. Scala's attempt to specialize Tuple2 Int argument does more harm than good!
 Tip: Use polymorphism sparingly, use the same order / patterns for nested & wrappered code, and
keep call sites homogeneous.

Replacing the object serializer

Writing a Spark-friendly object serializer
 Expand the list of specialist serialized form
– Having custom write/read object methods allows for reduced time in reflection and
smaller on-wire payloads.
– Types such as Tuple and Some given special treatment in the serializer
 Reduce the payload size further using variable length encoding of primitive types.
– All objects are eventually decomposed into primitives
 Adaptive stack-based recursive serialization vs. state machine serialization
– Use the stack to track state wherever possible, but fall back to state machine for deeply
nested objects (e.g. big RDDs)
 Sharing object representation within the serialized stream to reduce payload
– But may be defeated if supportsRelocationOfSerializedObjects required
 Special replacement of deserialization calls to avoid stack-walking to find class loader
context
– Optimization in JIT to circumvent some regular calls to more efficient versions
 Tip: These are opaque to the application, no special patterns required.

Faster IO – networking and storage

Remote Direct Memory Access (RDMA) Networking
Spark VM
Buffer
Off
Heap
Buffer
Spark VM
Buffer
Off
Heap
Buffer
Ether/IB
SwitchRDMA NIC/HCA RDMA NIC/HCA
OS OS
DMA DMA
(Z-Copy) (Z-Copy)
(B-Copy)(B-Copy)
Acronyms:
Z-Copy – Zero Copy
B-Copy – Buffer Copy
IB – InfiniBand
Ether - Ethernet
NIC – Network Interface Card
HCA – Host Control Adapter
●
Low-latency, high-throughput networking
●
Direct 'application to application' memory pointer exchange between remote hosts
●
Off-load network processing to RDMA NIC/HCA – OS/Kernel Bypass (zero-copy)
●
Introduces new IO characteristics that can influence the Spark transfer plan
Spark node #1 Spark node #2

Faster network IO with RDMA-enabled Spark
15
New dynamic transfer plan that adapts to the load and
responsiveness of the remote hosts.
New “RDMA” shuffle IO mode with lower latency and
higher throughput.
JVM-agnostic
IBM JVM only
JVM-agnostic
IBM JVM only
IBM JVM only
Block manipulation (i.e., RDD partitions)
High-level API
JVM-agnostic working prototype
with RDMA

Faster storage with POWER CAPI/Flash
 POWER8 architecture offers a 40Tb Flash drive attached
via Coherent Accelerator Processor Interface (CAPI)
– Provides simple coherent block IO APIs
– No file system overhead
 Power Service Layer (PSL)
– Performs Address Translations
– Maintains Cache
– Simple, but powerful interface to the Accelerator unit
 Coherent Accelerator Processor Proxy (CAPP)
– Maintains directory of cache lines held by Accelerator
– Snoops PowerBus on behalf of Accelerator


Faster disk IO with CAPI/Flash-enabled Spark
 When under memory pressure, Spark spills RDDs to disk.
– Happens in ExternalAppendOnlyMap and ExternalSorter
 We have modified Spark to spill to the high-bandwidth, coherently-attached
Flash device instead.
– Replacement for DiskBlockManager
– New FlashBlockManager handles spill to/from flash
 Making this pluggable requires some further abstraction in Spark:
– Spill code assumes using disks, and depends on DiskBlockManger
– We are spilling without using a file system layer
 Dramatically improves performance of executors under memory pressure.
 Allows to reach similar performance with much less memory (denser
deployments).
IBM Flash System 840Power8 + CAPI

Data Access Acceleration & auto SIMD

Data Access Accelerators (DAA)
 Provides JIT-recognized operations directly to/from byte arrays.
– Think of it as org.apache.spark.unsafe.Platform APIs on steroids!
 Optimized data conversion, arithmetic, etc on native format data records and types.
– Avoids object creation, initialization, data copying, abstraction, boxing, etc
– Range of operations on Packed Decimals; BigDecimal and BigInteger conversions from and
to external decimal and Unicode decimal types; marshalling and unmarshalling primitive
types (short, int, long, float, and double) to and from byte arrays
– e.g.
Benefits:
Expose hardware acceleration in a platform and JVM-neutral manner (2 – 100x speed-up)
Can provide significant speed-up to record parsing, data marshalling and inter-language
communication
 Tip: Don't be limited to current set of sun.misc.Unsafe APIs. Use existing DAA APIs, or
write equivalent places in Spark we can recognize.

Data Access Accelerator – example
Problem: there is no intrinsic representation of a “packed decimal” in Java
Solution: trivially add two packed decimals by converting to BigDecimal:
Step 1: convert from byte[] to BigDecimal
Step 2 : add BigDecimals together
Step 3: convert BigDecimal result back to byte[]
IBM Java provides a helper method to do this operation for you:
Old Approach:
byte[] addPacked(byte[] a, byte[] b) {
BigDecimal a_bd = convertPackedToBd(a[]);
BigDecimal b_bd = convertPackedToBd(b[]);
a_bd.add(b_bd);
return (convertBDtoPacked(a_bd));
}
BigDecimal convertPackedToBd(byte[] myBytes) {
… ~30 lines …
}
New Approach:
byte[] addPacked(byte[] a, byte[] b) {
DAA.addPacked(a[], b[]);
return (a[]);
}
e.g. you wish to add two packed decimal numbers
p.s. zSeries hardware has packed decimal machine instructions, e.g. AP (AddPacked)
On IBM Java 7 Release 1, “DAA.addPacked(a,b)” is compiled to this one instruction!
http://www01.ibm.com/support/knowledgecenter/SSYKE2_8.0.0/com.ibm.java.lnx.80.doc/user/daa.html

Automatic Vector Processing: SIMD – Single Instruction Multiple Data
The IBM JVM can operate on multiple data-elements (vectors) simultaneously
 Can offer dramatic speed-up to data-parallel operations (matrix ops, string processing, etc)
21
Vector registers are 128-bits wide and can be used
to operate concurrently on:
• Two 64-bit floating point values
• Four 32-bit floating point values
• Sixteen 8-bit characters
• Two 64-bit integer values
• etc

Offloading tasks to graphics co-processors

GPU-enabled array sort method
IBM Power 8 with Nvidia K40m GPU
 Some Arrays.sort() methods will offload work to GPUs today
– e.g. sorting large arrays of ints

GPU optimization of Lambda expressions
Speed-up factor when run on a GPU enabled host
IBM Power 8 with Nvidia K40m GPU
0.00
0.01
0.10
1.00
10.00
100.00
1000.00
auto-SIMD parallel forEach on CPU
parallel forEach on GPU
matrix size
The JIT can recognize parallel stream
code, and automatically compile down to
the GPU.

GPU off-loading of high-level ML functions

Summary
 All the changes are being made in the Java runtime or being pushed out to the Spark
community.
 We are focused on Core performance to get a multiplier up the Spark stack.
– More efficient code, more efficient memory usage/spilling, more efficient serialization &
networking.
– Need to take a look at bytecode codegen changes coming in via Tungsten
 There are hardware and software technologies we can bring to the party.
– We can tune the stack from hardware to high level structures for running Spark.
 Spark and Scala developers can help themselves by their style of coding.
 There is lots more stuff I don't have time to talk about, like GC optimizations, object layout,
monitoring VM/Spark events, hardware compression, security, etc. etc.

Five cool ways the JVM can run Apache Spark faster

More Related Content

What's hot (20)

Viewers also liked (10)

Similar to Five cool ways the JVM can run Apache Spark faster (20)

More from Tim Ellison (9)

Recently uploaded (20)

Five cool ways the JVM can run Apache Spark faster