Actor Patterns in Distributed Systems

using Dynamic Provisioning and SLA Management

Paul McGrath and Shaine Ismail

Javawug BOF-56 PSMI Limited

Introduction

• Biographies

• Actor patterns in Distributed Systems

• Enterprise IT Challenges

• Trends and Needs

• Fault tolerance / SLAs

• Putting it all together - a demo

• Questions

• IT Software Developer (19 years)

• Wide range of business domains

• Finance/Banking since 1995

• Expertise in large scale systems

design/implementation and delivery

• Founder/organiser of London Cloud Computing

Meetup group

Biography: Paul McGrath

• Java Developer for 10 Years.

• JPMorgan, BT, Lehman Brothers, Swiss Re.

• Experienced in delivering hi-performance, low

maintenance solutions.

Biography: Shaine Ismail

Kylie Minogue website

Failures to Scale - examples

iPhone launch

London Stock Exchange

.

London Stock Exchange (again)

Northern Rock

• Lack of scalability

• Failure to adapt to short-term volume spikes

• Operational risk

• Reputational risk → Company failure

Common issues

Enterprise IT Challenges

•Fast changing business requirements

•Short term project horizons

•Demanding customers

•Small budgets - do more with less

•Reactive management

•Hardware trends

•Rising energy costs

IT Trends

•Clock speeds levelling off

•More execution cores per machine

•More machines available - more resources

•Migration towards commodity hardware

•Heterogeneous execution environments

•Offsite Clouds/Machine clusters

•Green agenda

Enterprise Needs

IT Execution model that scales across multiple cores/multiple heterogeneous machines → concurrent programming

 "For concurrent programming to become

mainstream, we must discard threads as aprogramming model" 

- Ed Lee, The Problem with ThreadsBerkeley CS Tech Report

Actors - a definition

"Actors form the universal primitives of concurrent digital computation - in response to a message that it receives, an actor can make local decisions, create more actors, send more messages, and determine how to respond to the next message received" 

 (wikipedia 2010)

Actors

Actors encapsulate •     state•     instructions•     and execution

Actors provide architectural bind points whose execution can be monitored and to which we can associate Service Level Agreements.

Actors need:  Direction

– Where they should execute 

– How many copies should be executed

Roles

– Bespoke functional responsibility

Scripts

–  Exact, repeatable behaviour to message receipt

Chameleon

implements Actors in a distributed system - allowing dynamic provisioning of actors into a system, and SLA management of said actors.

Provides dynamic load balancing and active workload scaling.

→ Adaptive Execution Fabric

Chameleon

Directed Acyclic Graph Execution  

  

Modelled as a Workflow class – with split, join and block directives 

Chameleon

A demo....

Chameleon - under the hood

Java  Jini /JavaSpaces Rio

• Not a complete solution

• Potential integration points

Data grids (Coherence, EHCache etc), Rules Engines

(Drools, Esper)

• Potential use cases

Batch processing, ETL, STP, Workflow, Load

Balancing

Chameleon – the whole solution ?

Questions?