Cloud Computing and the NetBeans IDE Enable the Army Research Laboratory’s Next-Generation Simulation System. Ron Bowers Army Research Laboratory Dennis Reedy Elastic Grid LLC. 3 June 2009

Agenda Where we work and what we do Overview of the architecture Why and how we are using Cloud Computing Next steps and follow up

Where we work and what we do

Overview of the architecture

Why and how we are using Cloud Computing

Next steps and follow up

The Army Research Laboratory The Army Research Laboratory (ARL) is the Army's corporate basic and applied research laboratory. Our mission is to provide innovative science, technology, and analysis to enable full-spectrum operations. We represent the Survivability/Lethality Analysis Directorate (SLAD) of ARL.

The Army Research Laboratory (ARL) is the Army's corporate basic and applied research laboratory. Our mission is to provide innovative science, technology, and analysis to enable full-spectrum operations.

We represent the Survivability/Lethality Analysis Directorate (SLAD) of ARL.

We have some experience with computers...

Ensure that US personnel and equipment… SLAD Mission Nuclear, Biological and Chemical Attack Ballistic Threats Electronic Warfare Information Warfare … survive and function effectively in hostile circumstances

Ensure that US personnel and equipment…

SLAD performs both experimentation and modeling

SLAD Ballistic Vulnerability/Lethality (V/L) Modeling SLAD’s primary tool for performing ballistic V/L analysis is MUVES. MUVES development began in 1984. The current version (2.16) is a single-threaded C application. We are currently developing MUVES 3, which is an all-new replacement system.

SLAD’s primary tool for performing ballistic V/L analysis is MUVES.

MUVES development began in 1984.

The current version (2.16) is a single-threaded C application.

We are currently developing MUVES 3, which is an all-new replacement system.

Program Focus Provide the next generation of simulation system for the V/L analyst community Mostly Java. Dynamic distributed and service-oriented. Will support over 100 concurrent users. Incorporates a computational grid, parallelized system that distributes tasks and computes results that are graphically displayed. Will operate in both “batch” and interactive modes.

Provide the next generation of simulation system for the V/L analyst community

Mostly Java.

Dynamic distributed and service-oriented.

Will support over 100 concurrent users.

Incorporates a computational grid, parallelized system that distributes tasks and computes results that are graphically displayed.

Will operate in both “batch” and interactive modes.

Interesting Challenges We have few servers, but many powerful workstations. Architecture must exploit analyst community machines Share CPU, memory and disk Heterogeneous deployment environment Required application assets vary Need real-time provisioning of application assets Must be able to route functionality to machines that are best capable of executing tasks/functions Must be able to scale on demand based on real time need and use of the system Legacy of performance issues and nightmares

We have few servers, but many powerful workstations.

Architecture must exploit analyst community machines

Share CPU, memory and disk

Heterogeneous deployment environment

Required application assets vary

Need real-time provisioning of application assets

Must be able to route functionality to machines that are best capable of executing tasks/functions

Must be able to scale on demand based on real time need and use of the system

Legacy of performance issues and nightmares

Solution Approach Choose technology that embraces dynamic distributed capabilities Craft a loosely coupled service oriented architecture that segments the system into functional roles Choose persistence technologies and approaches that allow for low latency and high concurrency Represent data as it moves through the system In-flight (hot in-memory), Swap, Long Term, Archived Keep Disk I/O out of the main stream processing

Choose technology that embraces dynamic distributed capabilities

Craft a loosely coupled service oriented architecture that segments the system into functional roles

Choose persistence technologies and approaches that allow for low latency and high concurrency

Represent data as it moves through the system

In-flight (hot in-memory), Swap, Long Term, Archived

Keep Disk I/O out of the main stream processing

What’s Underneath Application Infrastructure Quality of Service Monitoring and Management Domain-specific Services and Algorithms Dynamic Container Persistence Management

What’s Underneath Application Infrastructure Quality of Service Monitoring and Management Domain-specific Services and Algorithms Dynamic Container Persistence Management JavaSpace Apache Active MQ Apache Derby

What’s Underneath Application Infrastructure Quality of Service Monitoring and Management Domain-specific Services and Algorithms Dynamic Container Persistence Management Rio

What’s Underneath Application Infrastructure Quality of Service Monitoring and Management Domain-specific Services and Algorithms Dynamic Container Persistence Management Gomez

Gomez Was established as a prototype for MUVES 3 architectural enhancements. Now forms the basis of the MUVES 3 service-oriented architecture (SOA). Includes all of the non-sensitive services used by MUVES 3. Is an open source project (LGPLv3).

Was established as a prototype for MUVES 3 architectural enhancements.

Now forms the basis of the MUVES 3 service-oriented architecture (SOA).

Includes all of the non-sensitive services used by MUVES 3.

Is an open source project (LGPLv3).

What’s Underneath Application Infrastructure Quality of Service Monitoring and Management Domain-specific Services and Algorithms Dynamic Container Persistence Management

MUVES 3 System Organization Attachment point for clients. Monitors system load. Controls job submission. Executes analysis jobs. Stores analysis results. Gateway Sim Persistence Client MUVES 3 UI

Attachment point for clients.

Monitors system load.

Controls job submission.

Executes analysis

jobs.

Stores analysis

results.

MUVES 3 UI The MUVES 3 UI is built on the NetBeans Platform. Used for: Input preparation Team collaboration Job submittal and monitoring. Result visualization Components that interact with back-end services are developed in Gomez and used in the MUVES 3 UI.

The MUVES 3 UI is built on the NetBeans Platform.

Used for:

Input preparation

Team collaboration

Job submittal and monitoring.

Result visualization

Components that interact with back-end services are developed in Gomez and used in the MUVES 3 UI.

Our NetBeans Experience Favorite things Swing-based Fast form development Easy deployment via JNLP Extremely well supported by the community Biggest issue Integrating libraries that are updated frequently

Favorite things

Swing-based

Fast form development

Easy deployment via JNLP

Extremely well supported by the community

Biggest issue

Integrating libraries that are updated frequently

MUVES 3 Execution Client Gateway Persistence Sim Pool Busy Sim Pool Sim Pool

MUVES 3 Execution Client Gateway Select Sim Pool Persistence Sim Pool Busy Sim Pool Sim Pool Submit job

MUVES 3 Execution Client Gateway Sim Pool Persistence Worker Worker Worker Worker Job Monitor Task Space

MUVES 3 Execution Client Gateway Sim Pool Persistence Ray Tracer Personnel Vulnerability Specialized Physics Vehicle Performance Worker Worker Worker Worker Job Monitor Task Space Deploy additional services

MUVES 3 Execution Client Gateway Sim Pool Persistence Ray Tracer Personnel Vulnerability Specialized Physics Vehicle Performance Worker Worker Worker Worker Job Monitor Task Space Submit job Store results Visualize Results

Dynamic Clustering Client Proxy Service Interface Service Available strategies: Fail-Over – Uses one service unless that service goes away. Round-Robin – Iterates over all discovered services. Utilization – Like round-robin, but ignores services that are low on system resources. Service Service Rio Association declaration Service discovery Service injection Service Selection Strategy Proxy injection

Available strategies:

Fail-Over – Uses one service unless that service goes away.

Round-Robin – Iterates over all discovered services.

Utilization – Like round-robin, but ignores services that are low on system resources.

The Persistence Meta-service Stores the results of analyses. Consists of four layers: In-memory cache (distributed JavaSpace), Swap, (Apache Active MQ) Long-term storage (Apache Derby), and Archive (Hibernate+Oracle). Layers are implemented using dynamic clustering. Supports data life-cycle management Data storage is leased, and leases can expire. When the cache fills, results are moved to swap. Must work for the next 20 years.

Stores the results of analyses.

Consists of four layers:

In-memory cache (distributed JavaSpace),

Swap, (Apache Active MQ)

Long-term storage (Apache Derby), and

Archive (Hibernate+Oracle).

Layers are implemented using dynamic clustering.

Supports data life-cycle management

Data storage is leased, and leases can expire.

When the cache fills, results are moved to swap.

Must work for the next 20 years.

Cloud Computing Public Cloud: Obvious national security issues Private Cloud Conceptually already there Realtime provisioning of applications Dynamic allocation SLA based approach

Public Cloud:

Obvious national security issues

Private Cloud

Conceptually already there

Realtime provisioning of applications

Dynamic allocation

SLA based approach

Cloud Computing:Applicable for Testing? Goals Demonstrate the performance and scalability of MUVES 3 over dozens (or hundreds :-) ) of computers. Execute multiple integration tests concurrently. Issues Small number of computers available locally for testing. Difficult coordination issues due to Army security policies. Approach Test the MUVES 3 architecture (Gomez) on Amazon Elastic Compute Cloud (EC2)

Goals

Demonstrate the performance and scalability of MUVES 3 over dozens (or hundreds :-) ) of computers.

Execute multiple integration tests concurrently.

Issues

Small number of computers available locally for testing.

Difficult coordination issues due to Army security policies.

Approach

Test the MUVES 3 architecture (Gomez) on Amazon Elastic Compute Cloud (EC2)

Cloud Adoption Challenges Getting approval :) Administrative burden We don’t want to build AMIs or go through the time to provision an entire stack every night. Minimize changes Avoid developing special code and testing framework for cloud deployment/orchestration. Ideally, transparently switch from LAN-based deployment to the cloud. We must preserve the dynamic distributed semantics architected in the system: Service selection strategies Dynamic discovery semantics Run multiple concurrent test cases and roll up test results.

Getting approval :)

Administrative burden

We don’t want to build AMIs or go through the time to provision an entire stack every night.

Minimize changes

Avoid developing special code and testing framework for cloud deployment/orchestration.

Ideally, transparently switch from LAN-based deployment to the cloud.

We must preserve the dynamic distributed semantics architected in the system:

Service selection strategies

Dynamic discovery semantics

Run multiple concurrent test cases and roll up test results.

Cloud Adoption Approach Use Elastic Grid (EG) Eases development and deployment of Java applications into the Cloud Provides automated management, fault detection, and scalability for the application Allows focus on development, not cloud infrastructure

Use Elastic Grid (EG)

Eases development and deployment of Java applications into the Cloud

Provides automated management, fault detection, and scalability for the application

Allows focus on development, not cloud infrastructure

Elastic Grid Overview Cloud Management Fabric Dynamically instantiate, monitor & manage application components SLA policy driven with strategies like service scalability, relocation, fault detection & recovery, etc. Cloud Virtualization Layer Abstracts specific Cloud Computing provider technology Allows portability across specific implementations You can deploy on Amazon EC2, Private LAN based Cloud; Eucalyptus, Sun Cloud and others soon

Cloud Management Fabric

Dynamically instantiate, monitor & manage application components

SLA policy driven with strategies like service scalability, relocation, fault detection & recovery, etc.

Cloud Virtualization Layer

Abstracts specific Cloud Computing provider technology

Allows portability across specific implementations

Cloud Activation & Deployment Groovy Client Application Monitors Application Agents 2 Create Clusters Download and deploy application resources Deploy 3 Upload JUnit test results 4 Build, Create Release, Upload Archive S3 1 Download Test results and post process 5

Cloud Activation & Deployment Test Cluster Test Cluster Test Cluster Groovy Client S3

Extending Continuous Integration Automated build and test, both unit and integration tests Extend this to include Continuous Deployment If CI passes, use Test Cloud Bursting to deploy, verify and validate system SVN Repo. Deploy system and run tests Retrieve and process test results Test Runner

Automated build and test, both unit and integration tests

Extend this to include Continuous Deployment

If CI passes, use Test Cloud Bursting to deploy, verify and validate system

Next Steps Lots to think about

Lots to think about

Summary Using the public cloud for scalability testing and verification a good choice Without Elastic Grid we would have a much more difficult experience (may not have done it) Looking to expand metrics gathering & collection Looking to incorporate NetBeans as a client for cloud visualization We are looking to make this part of our permanent development environment How to spread the gospel of cloud computing for DoD

Using the public cloud for scalability testing and verification a good choice

Without Elastic Grid we would have a much more difficult experience (may not have done it)

Looking to expand metrics gathering & collection

Looking to incorporate NetBeans as a client for cloud visualization

We are looking to make this part of our permanent development environment

How to spread the gospel of cloud computing for DoD

Dennis Reedy Elastic Grid LLC [email_address] Ron Bowers Army Research Laboratory [email_address]