36

Betting On Data Grids

Embed Size (px)

DESCRIPTION

From the Gaming Scalability event, June 2009 in London (http://gamingscalability.org). Dave Felcey from Oracle will give an overview of Oracle Coherence and releted technologies, like JRockit Real-Time JVM, and discuss how they are being used to address some of the challenges their gaming customers face. In the gaming industry real-time updates and resilience are key. Getting price changes to users by caching data in memory and pushing real-time changes to clients using Coherence can provides a competitive edge and attracts new customers. Increasingly holding data in-memory and using the real-time tools are the only way sites can meet user expectations. However, ensuring in-memory data is resilient under load is also crucial, to protect against costly outages at key times. Dave will discuss the technical details and approaches that can be used to meet these requirements.

Citation preview

Page 1: Betting On Data Grids
Page 2: Betting On Data Grids

<Insert Picture Here>

Betting on Data Grids

Dave Felcey

Oracle Sales Consulting

Page 3: Betting On Data Grids

Agenda

• Oracle High Performance Computing

• Oracle Coherence Architecture

• Gaming Industry Challenges

• Summary

Page 4: Betting On Data Grids

Oracle High Performance ComputingComprehensive and Best of Breed

• Oracle 11g WebLogic Server• Fastest Applicaton Server, delivering 7,311 SPECjAppServer2004

JOPS@Standard (jAppServer Operations per Second)

• Oracle JRockit Real-Time JVM• Fastest JVM, delivering 537,116 SPECjbb2005 bops/JVM p/s

• Oracle Complex Event Processing• Fraud detection, risk mitigation etc.

• Oracle 11g Database• Used by Betfair for performance and scalability and one of top 5

busiest databases in the world

• Oracle TimesTen In-Memeory Database• The Hong Kong Jockey Club uses TimesTen to perform very fast

fraud detection processing

• Oracle Identity Management (IdM)• Used by Shanda to manage ID of upto 2M concurrent users

Page 5: Betting On Data Grids

Oracle High Performance ComputingComprehensive and Best of Breed

Low Latency Predictable

JRockitJRockit RealReal--Time JVMTime JVM

Low Latency

Scalable

Resilient

Coherence Data GridCoherence Data Grid

Low Latency

Complex Event Complex Event ProcessingProcessing

Content Cache J2EE and Messaging

WebCacheWebCache WebLogicWebLogic ServerServer

Monitoring

Management Management ToolsTools

Scale OutCommodity Hardware

Oracle RACOracle RAC

SLA’sand QoS

Diagnostics

Provisioning

EQL

XML

Embedded

Transactional

Berkeley DBBerkeley DB

Fast

Low Latency SQL

TimesTenTimesTen

In-Memory

TuxedoTuxedo

Low Latency

TPM

Mature and

Proven

Page 6: Betting On Data Grids

Oracle CoherenceData Grid Uses

Caching

Applications request data from the Data Grid rather than

backend data sources

Analytics

Applications ask the Data Grid questions from simple queries to

advanced scenario modeling

Transactions

Data Grid acts as a transactional System of Record, hosting

data and business logic

Events

Automated processing based on event

Page 7: Betting On Data Grids

The Coherence Approach…

• Consensus is key

• Communication is more efficient (peer-to-peer)

• No outages for voting (no need – everyone is a peer)

• No SPoF, SPoB

• No need for broadcast traffic (yelling at each other)

• You can do many things once you have “consensus”.

Page 8: Betting On Data Grids

TCMP Provides the Foundations

Page 9: Betting On Data Grids

What is Coherence?

• Coherence (deployment perspective)

• Single Library*

• *Other libraries for integration (L2C, Spring…)

• Configurable implementations of standard Map interfaces

(called NamedCache’s)

• Standard Java Archive “JAR” for Java

• Standard Dynamically Linked Library “DLL” for .NET

connectivity (.Net 1.1 and 2.0)

• Standard DLL or .so for C++ clients

• No 3rd party dependencies!

• Minimal “invasion” on standard code*

• “RemoteException” free distributed computing

Page 10: Betting On Data Grids

Introduction to NamedCaches

• Developers use NamedCaches to manage data

• An composite interface which includes Map

• NamedCache

• Logically equivalent to a Database table

• Store related types of information (trades, orders, sessions)

• May be hundreds / thousands of per Application

• May be dynamically created

• May contain any data (no need to setup a schema)

• No restriction on types (homogeneous and heterogeneous)

• Not relational (but may be)

Page 11: Betting On Data Grids

Clustered Hello World!

public void main(String[] args) throws IOException {

NamedCache nc = CacheFactory.getCache(“test”);

nc.put(“key”, “Hello World”);

System.out.println(nc.get(“key”));

System.in.read(); //may throw exception

}

• Joins / Establishes a cluster

• Places an Entry (key, value) into the Cache “test” (notice no

configuration)

• Retrieves the Entry from the Cache.

• Displays it.

• “read” at the end to keep the application (and Cluster) from

terminating.

Page 12: Betting On Data Grids

Caching Strategies (schemes)Different cache implementations

• Local

• Local on-heap caching for non-clustered caching.

• Replicated

• Perfect for small, read-heavy caches.

• Partitioned

• True linear scalability for both read and write access. Data is automatically, dynamically and transparently partitioned across nodes. The distribution algorithm minimizes network traffic and avoids service pauses by incrementally shifting data.

• Near Cache

• Provides the performance of local caching with the scalability of distributed caching. Several different near-cache strategies provide varying tradeoffs between performance and synchronization guarantees.

Page 13: Betting On Data Grids

The Distributed Scheme - Get

Page 14: Betting On Data Grids

The Distributed Scheme - Put

Page 15: Betting On Data Grids

The Distributed Scheme - Failover

Page 16: Betting On Data Grids

The Near Scheme

• A composition of pluggable Front and Back schemes

• Provides L1 and L2 caching (cache of a cache)

• Why:

• Partitioned Topology may always go across the wire

• Need a local cache (L1) over the distributed scheme (L2)

• Best option for scalable performance!

• How:

• Configure ‘front’ and ‘back’ topologies

• Configurable Expiration Policies:

• LFU, LRU, Hybrid (LFU+LRU), Time-based, Never,

Pluggable

Page 17: Betting On Data Grids

The Near Scheme - Get

Page 18: Betting On Data Grids

Coherence*Extend

Page 19: Betting On Data Grids

WAN Topology

Page 20: Betting On Data Grids

Queries

• Filters applied in parallel (in the Grid)

• A large range of filters out-of-the-box:

All, Always, And, Any, Array, Between,

ContainsAll, ContainsAny, Contains, Equals,

GreaterEquals, Greater, In, InKeySet,

IsNotNull, IsNull, LessEquals, Less, Like,

Limit, Never, NotEquals, Not, Or…

Filter filter = new AndFilter(

new EqualsFilter("getTrader", traderId),

new EqualsFilter("getStatus", Status.OPEN));

Set setOpenTrades = mapTrades.entrySet(filter);

Page 21: Betting On Data Grids

Executing a query

Page 22: Betting On Data Grids

Real Time Events

• Maintain real time visibility into data changes

• Desktops

• The usual example is the “Trader desktop”

• Watch data change in near real time

• Typically a few milliseconds

• Servers

• Monitoring data to trigger additional processing

• Event Driven Architecture within the data grid

• Very wide-ranging set of use cases

• Not many common patterns of usage

Page 23: Betting On Data Grids

Continuous Query Cache

Coherence implements Continuous Query using a combination

of its data fabric parallel query capability and its real-time event-

filtering and streaming. The result is support for thousands of

client application instances, such as trading desktops. Using the

previous trading system example, it can be converted to a

Continuous Query with only one a single line of code changed

NamedCache mapTrades = ...

Filter filter = new AndFilter(new

EqualsFilter("getTrader", traderid),

new EqualsFilter("getStatus", Status.OPEN));

NamedCache mapOpenTrades = new

ContinuousQueryCache(mapTrades, filter);

Page 24: Betting On Data Grids

Transaction Management

• Explicit transaction management• Using the general pattern for pessimistic transactions is "lock

-> read -> write -> unlock". For optimistic transactions, the sequence is "read -> lock & validate -> write -> unlock".

• Implicit transaction management• Locking "by convention" – for example, requiring that all

acessors lock only the "parent" Order object. Doing this can reduce the scope of the lock from table-level to order-level, enabling far higher scalability

• Further transaction optimizations• Using EntryProcessors – sending the code to the data, so

that operations are queued and all locking is local. Operations must be idempotent.

Page 25: Betting On Data Grids

Cache ThroughReading ahead or on-demand

Page 26: Betting On Data Grids

Persisting DataWrite-through, write-behind, coalescing and batching

Page 27: Betting On Data Grids

HTTP Session Caching

Overview

• No code changes required to use

• Portlet state can be cached

• Built into WLS and WLP

Benefits

• Enables stateless middle tier

• Better hardware utilization

• Simpler network infrastructure

• Facilitates modular application improvements

• Scales out middle tier

Page 28: Betting On Data Grids

Serialization Portable Object Format (POF)

• Benefits

• Can store more data

• Can read/write and move data faster

Coherence Compression Test Results

867

309 322

186

0

100

200

300

400

500

600

700

800

900

1000

Java ExternalizationLite XMLBean POF

Byte

s

Java

ExternalizationLite

XMLBean

POF

Serialization

De-serialization

10078

1625 2070

12342360

484734

547

0

2000

4000

6000

8000

10000

12000

Time (ms)

Serialization Mechanisum

Coherence Serialization Test Results

Serialization

De-serialization

5x Smaller 10x Faster De-Serialization

Page 29: Betting On Data Grids

Coherence IncubatorPatterns

• Pre-built examples

• Used in production systems

• Thoroughly tested

• Extensible

• Optimised

• Incorporate best practice

Page 30: Betting On Data Grids

Gaming Challenges

• Extreme scalability 500k+ users

• Reliability. Outages damage reputation and can cost £100k+ p/hr

• Flexibility. Enable products to be quickly brought to

market

Page 31: Betting On Data Grids

Extreme Scalability

• Scaling Users

• 100k – 1M online users

• Asynchronously update database so reduce latency, open

connections etc.

• Scaling Transactions and Processing

• Betfair

• INCERNO processed 5k TPS in simulation tests with no

discernable deterioration in performance or reliability.

• Scaling Data Capacity, >100 GB

• Off-heap storage option in release 3.5

• Potential storage limit now > TB

Page 32: Betting On Data Grids

Extreme Reliability

• Non-Stop running

• 2 years+ continuous running

• Withstand database or link replication failure

• Queue requests

• Failure of multiple servers

• No ‘Single Point Of Failure’

• Processing (as well as data) failover

Page 33: Betting On Data Grids

Extreme Flexibility

• Native Java, C++ and .NET clients

• Simple Map and IDictionary API

• Simple to install

• Pre-built examples (Incubator Projects)

• Seamless HTTP Session integration for J2EE and .NET

• Support of Hibernate, JPA and Spring

Support

• Active forums and SIG’s

• Well documented

Page 34: Betting On Data Grids

<Insert Picture Here>

Summary

• Coherence™ is the leading product for high

performance distributed in-memory data services• Proven technology, 100+ customers and 1500+

production systems

• Offers a unique combination of features

• Coherence™ is easy to use and delivers

data performance, scalability and reliability

Page 35: Betting On Data Grids
Page 36: Betting On Data Grids