| By Sanat Vij | Article Rating: |
|
| March 15, 2012 10:00 AM EDT | Reads: |
4,877 |
At a time where the clock speeds of processors have been stable over the past couple of years, and Moore's Law is instead being applied by increasing the number of processor cores, it is getting more important for applications to use concurrent processing to reduce run/response times, as the time slicing routine via increased clock speed will no longer be available to bail out slow running programs.
Carl Hewitt proposed the Actor Model in 1973 as a way to implement unbounded nondeterminism in concurrent processing. In many ways this model was influenced by the packet switching mechanism, for example, no synchronous handshake between sender and receiver, inherently concurrent message passing, messages may not arrive in the order they were sent, addresses are stored in messages, etc.
The main difference between this model and most other parallel processing systems is that it uses message passing instead of shared variables for communication between concurrent processes. Using shared memory to communicate between concurrent processes requires the application of some form of locking mechanism to coordinate between threads, which may give rise to live locks, deadlocks, race conditions and starvation.
Actors are the location transparent primitives that form the basis of the actor model. Each actor is associated with a mailbox (which is a queue with multiple producers and a single consumer) where it receives and buffers messages, and a behavior that is executed as a reaction to a message received. The messages are immutable and may be passed between actors synchronously or asynchronously depending on the type of operation being invoked. In response to a message that it receives, an actor can make local decisions, create more actors, send more messages, and designate how to respond to the next message received. Actors never share state and thus don't need to compete for locks for access to shared data.
The actor model first rose to fame in the language Erlang, designed by Ericcson in 1986. It has since been implemented in many next-generation languages on the JVM such as Scala, Groovy and Fantom. It is the simplicity of usage provided via a higher level of abstraction that makes the actor model easier to implement and reason about.
It's now possible to implement the actor model in Java, thanks to the growing number of third-party concurrency libraries advertising this feature. Akka is one such library, written in Scala, that uses the Actor model to simplify writing fault-tolerant, highly scalable applications in both Java and Scala.
Implementation
Using Akka, we shall attempt to create a concurrent processing system for loan request processing in a bank as can be seen in the Figure 1.

Figure 1
The system consists of four actors:
- The front desk - which shall receive loan requests from the customers and send them to the back office for processing. It shall also maintain the statistics of the number of loans accepted/rejected and print a report detailing the same on being asked to do so.
- The back office - which shall sort the loan requests into personal loans and home loans, and send them to the corresponding accountant for approval/rejection.
- The personal loan accountant - who shall process personal loan requests, including approving/rejecting the requests, carrying out credit history checks and calculating the rate of interest.
- The home loan accountant - who shall process home loan requests, including approving/rejecting the requests, carrying out credit history checks and calculating the rate of interest.
To get started, download the version 2.0 of Akka for Java from http://akka.io/downloads and add the jars present in ‘akka-2.0-RC4\lib' to the classpath.
Next, create a new Java class "Bank.java" and add the following import statements :
import akka.actor.ActorRef;
import akka.actor.ActorSystem;
import akka.actor.Props;
import akka.actor.UntypedActor;
import akka.routing.RoundRobinRouter;
Now, create a few static nested classes under ‘Bank' that will act as messages (DTOs) and be passed to actors.
1. ‘LoanRequest' will contain the following elements and their corresponding getters/setters:
int requestedLoan;
int accountBalance;
2. ‘PersonalLoanRequest' will extend ‘LoanRequest' and contain the following element and its corresponding getter:
final static String type="Personal";
3. ‘HomeLoanRequest' will extend ‘LoanRequest' and contain the following element and its corresponding getter:
final static String type="Home";
4. ‘LoanReply' will contain the following elements and their corresponding getters/setters:
String type;
boolean approved;
int rate;
Next, create a static nested class ‘PersonalLoanAccountant' under ‘Bank'.
public static class PersonalLoanAccountant extends UntypedActor {
public int rateCaluclation(int requestedLoan, int accountBalance) {
if (accountBalance/requestedLoan>=2)
return 5;
else
return 6;
}
public void checkCreditHistory() {
for (int i=0; i<1000; i++) {
continue ;
}
}
public void onReceive(Object message) {
if (message instanceof PersonalLoanRequest) {
PersonalLoanRequest request=PersonalLoanRequest.class.cast(message);
LoanReply reply = new LoanReply();
reply.setType(request.getType());
if (request.getRequestedLoan()<request.getaccountBalance()) {
reply.setApproved(true);
reply.setRate(rateCaluclation(request.getRequestedLoan(), request.getaccountBalance()));
checkCreditHistory();
}
getSender().tell(reply);
} else {
unhandled(message);
}
}
}
The above class serves as an actor in the system. It extends the ‘UntypedActor' base class provided by Akka and must define its ‘onReceive‘ method. This method acts as the mailbox and receives messages from other actors (or non-actors) in the system. If the message received is of type 'PersonalLoanRequest', then it can be processed by approving/rejecting the loan request, setting the rate of interest and checking the requestor's credit history. Once the processing is complete, the requestor uses the sender's reference (which is embedded in the message) to send the reply (LoanReply) to the requestor via the ‘getSender().tell()' method.
Now, create a similar static nested class ‘HomeLoanAccountant' under ‘Bank'
public static class HomeLoanAccountant extends UntypedActor {
public int rateCaluclation(int requestedLoan, int accountBalance) {
if (accountBalance/requestedLoan>=2)
return 7;
else
return 8;
}
public void checkCreditHistory() {
for (int i=0; i<2000; i++) {
continue ;
}
}
public void onReceive(Object message) {
if (message instanceof HomeLoanRequest) {
HomeLoanRequest request=HomeLoanRequest.class.cast(message);
LoanReply reply = new LoanReply();
reply.setType(request.getType());
if (request.getRequestedLoan()<request.getaccountBalance()) {
reply.setApproved(true);
reply.setRate(rateCaluclation(request.getRequestedLoan(), request.getaccountBalance()));
checkCreditHistory();
}
getSender().tell(reply);
} else {
unhandled(message);
}
}
}
Now, create a static nested class ‘BackOffice' under ‘Bank':
public static class BackOffice extends UntypedActor {
ActorRef personalLoanAccountant=getContext().actorOf(new Props(PersonalLoanAccountant.class).withRouter(new RoundRobinRouter(2)));
ActorRef homeLoanAccountant=getContext().actorOf(new Props(HomeLoanAccountant.class).withRouter(new RoundRobinRouter(2)));
public void onReceive(Object message) {
if (message instanceof PersonalLoanRequest) {
personalLoanAccountant.forward(message, getContext());
} else if (message instanceof HomeLoanRequest) {
homeLoanAccountant.forward(message, getContext());
} else {
unhandled(message);
}
}
}
The above class serves as an actor in the system and is purely used to route the incoming messages to PersonalLoanAccountant or HomeLoanAccountant, based on the type of message received. It defines actor references ‘personalLoanAccountant' and ‘homeLoanAccountant' to ‘PersonalLoanAccountant.class' and ‘HomeLoanAccountant.class', respectively. Each of these references initiates two instances of the actor it refers to and attaches a round-robin router to cycle through the actor instances. The ‘onReceive' method checks the type of the message received and forwards the message to either ‘PersonalLoanAccountant' or ‘HomeLoanAccountant' based on the message type. The ‘forward()' method helps ensure that the reference of the original sender is maintained in the message, so the receiver of the message (‘PersonalLoanAccountant' or ‘HomeLoanAccountant') can directly reply back to the message's original sender.
Now, create a static nested class ‘FrontDesk' under ‘Bank':
public static class FrontDesk extends UntypedActor {
int approvedPersonalLoans=0;
int approvedHomeLoans=0;
int rejectedPersonalLoans=0;
int rejectedHomeLoans=0;
ActorRef backOffice=getContext().actorOf( new Props(BackOffice.class), "backOffice");
public void maintainLoanApprovalStats(Object message) {
LoanReply reply = LoanReply.class.cast(message);
if (reply.isApproved()) {
System.out.println(reply.getType()+" Loan Approved"+" at "+reply.getRate()+"% interest.");
if (reply.getType().equals("Personal"))
++approvedPersonalLoans;
else if(reply.getType().equals("Home"))
++approvedHomeLoans;
} else {
System.out.println(reply.getType()+" Loan Rejected");
if (reply.getType().equals("Personal"))
++rejectedPersonalLoans;
else if(reply.getType().equals("Home"))
++rejectedHomeLoans;
}
}
public void printLoanApprovalStats() {
System.out.println("--- REPORT ---");
System.out.println("Personal Loans Approved : "+approvedPersonalLoans);
System.out.println("Home Loans Approved : "+approvedHomeLoans);
System.out.println("Personal Loans Rejected : "+rejectedPersonalLoans);
System.out.println("Home Loans Rejected : "+rejectedHomeLoans);
}
public void onReceive(Object message) {
if (message instanceof LoanRequest) {
backOffice.tell(message, getSelf());
} else if (message instanceof LoanReply) {
maintainLoanApprovalStats(message);
} else if(message instanceof String && message.equals("printLoanApprovalStats")) {
printLoanApprovalStats();
getContext().stop(getSelf());
} else {
unhandled(message);
}
}
}
The above class serves as the final actor in the system. It creates a reference ‘backOffice' to the actor ‘BackOffice.class'. If the message received is of type ‘LoanRequest', it sends the message to ‘BackOffice' via the method ‘backOffice.tell()', which takes a message and the reference to the sender (acquired through the method ‘getSelf()') as arguments. If the message received is of type ‘LoanReply', it updates the counters to maintain the approved/rejected counts. If the message received is "printLoanApprovalStats", it prints the stats stored in the counters, and then proceeds to stop itself via the method ‘getContext().stop(getSelf())'. The actors follow a pattern of supervisor hierarchy, and thus this command trickles down the hierarchy chain and stops all four actors in the system.
Finally, write a few methods under ‘Bank' to submit requests to the ‘FrontOffice':
public static void main(String[] args) throws InterruptedException {
ActorSystem system = ActorSystem.create("bankSystem");
ActorRef frontDesk = system.actorOf(new Props(FrontDesk.class), "frontDesk");
submitLoanRequests (frontDesk);
Thread.sleep(1000);
printLoanApprovalStats (frontDesk);
system.shutdown();
}
public static PersonalLoanRequest getPersonalLoanRequest() {
int min=10000; int max=50000;
int amount=min + (int)(Math.random() * ((max - min) + 1));
int balance=min + (int)(Math.random() * ((max - min) + 1));
return (new Bank()).new PersonalLoanRequest(amount, balance);
}
public static HomeLoanRequest getHomeLoanRequest() {
int min=50000; int max=90000;
int amount=min + (int)(Math.random() * ((max - min) + 1));
int balance=min + (int)(Math.random() * ((max - min) + 1));
return (new Bank()).new HomeLoanRequest(amount, balance);
}
public static void submitLoanRequests(ActorRef frontDesk) {
for (int i=0;i<1000;i++) {
frontDesk.tell(getPersonalLoanRequest());
frontDesk.tell(getHomeLoanRequest());
}
}
public static void printLoanApprovalStats(ActorRef frontDesk) {
frontDesk.tell("printLoanApprovalStats");
}
The method ‘main' creates an actor system ‘system' using the method ‘ActorSystem.create()'. It then creates a reference ‘frontDesk' to the actor ‘FrontDesk.class'. It uses this reference to send a 1000 requests each of types ‘PersonalLoanRequest' and ‘HomeLoanRequest' to ‘FrontDesk'. It then sleeps for a second, following which it sends the message "printLoanApprovalStats" to ‘FrontDesk'. Once done, it shuts down the actor system via the method ‘system.shutdown()'.
Conclusion
Running the above code will create a loan request processing system with concurrent processing capabilities, and all this without using a single synchronize/lock pattern. Moreover, the code doesn't need one to go into the low-level semantics of the JVM threading mechanism or use the complex ‘java.util.concurrent' package. This mechanism of concurrent processing using the Actor Model is truly more robust as multiple concurrent processes can communicate with each other without needing to use shared state variables.
Published March 15, 2012 Reads 4,877
Copyright © 2012 SYS-CON Media, Inc. — All Rights Reserved.
Syndicated stories and blog feeds, all rights reserved by the author.
More Stories By Sanat Vij
Sanat Vij is a professional software engineer currently working at CenturyLink. He has vast experience in developing high availability applications, configuring application servers, JVM profiling and memory management. He specializes in performance tuning of applications, reducing response times, and increasing stability.
- Cloud People: A Who's Who of Cloud Computing
- New Relic Q1 2013 Blazes Past Growth Targets and Reaches 40,000 Active Customer Accounts
- Cloud Expo New York: Delivering Digital Marketing on the Cloud
- Cloudant to Exhibit at Cloud Expo & Big Data Expo New York
- Cloud Expo New York: Rethink IT and Reinvent Business with IBM SmartCloud
- The Accessibility of the Cloud
- Learn How To Use Google Apps Script
- Cloud Expo New York: Basics of SSD Technology and Its Use in Cloud
- Cloud Expo NY: Best Practices for Delivering Oracle Database as a Service
- Cloud Expo New York: Real-Time Analytics Using an In-Memory Data Grid
- Cloud Expo New York: The Big Challenge of Big Data & Hadoop Integration
- Measuring the Business Value of Cloud Computing
- Cloud People: A Who's Who of Cloud Computing
- Cloud Expo New York: Best CIO Practices Shared from SHI’s Customers
- Examining the True Cost of Big Data
- Cloud Expo New York: How to Use Google Apps Script
- New Relic Q1 2013 Blazes Past Growth Targets and Reaches 40,000 Active Customer Accounts
- Software Defined Networking – A Paradigm Shift
- Cloud Expo New York: Why Big Data Is Really About Small Data
- Cloud Expo New York: Delivering Digital Marketing on the Cloud
- Small Cancers, Big Data, and a Life Examined
- Cloud Expo New York: Requirements of a Cloud Database
- Cloudant to Exhibit at Cloud Expo & Big Data Expo New York
- Cloud Expo New York: Rethink IT and Reinvent Business with IBM SmartCloud
- A Cup of AJAX? Nay, Just Regular Java Please
- Java Developer's Journal Exclusive: 2006 "JDJ Editors' Choice" Awards
- JavaServer Faces (JSF) vs Struts
- The i-Technology Right Stuff
- Rich Internet Applications with Adobe Flex 2 and Java
- Java vs C++ "Shootout" Revisited
- Bean-Managed Persistence Using a Proxy List
- Reporting Made Easy with JasperReports and Hibernate
- Creating a Pet Store Application with JavaServer Faces, Spring, and Hibernate
- Why Do 'Cool Kids' Choose Ruby or PHP to Build Websites Instead of Java?
- What's New in Eclipse?
- Where Are RIA Technologies Headed in 2008?
























