Advanced Decision Architectures Collaborative Technology Alliance Agile Computing and its Applications to Tactical Army Environments Niranjan Suri, Marco Carvalho, Danilo Ansaloni, Marco Arguedas, Giacomo Benincasa, Erika Bevegn, Jeffrey M. Bradshaw, Maggie Breedy, Massimiliano Marcon, Raffaele Quitadamo, Matteo Rebeschini, Mauro Tortonesi Steve Choy, Jesse Kovach, Larry Tokarcik, Robert Winkler U.S. Army Research Laboratory Advanced Decision Architectures Collaborative Technology Alliance Outline Agile Computing Overview Mockets Communications Library Group Manager Discovery Service AgServe Service-oriented Architecture FlexFeed Publish-Subscribe System DisService Information Dissemination System Conclusions Advanced Decision Architectures Collaborative Technology Alliance Agile Computing Overview Both a Metaphor and an Approach to Distributed Information Systems Design Systems to be Opportunistic in discovering, manipulating, and exploiting computing and communication resources Quick in reacting and adapting to changes in the environment Able to take advantage of the wiggle room available in the system Assist disadvantaged users / systems by exploiting resource rich nodes Advanced Decision Architectures Collaborative Technology Alliance Measuring Agility Agility may be measured using How quickly a new resource can be utilized (Exploit Time) How long a resource needs to be available to be effectively utilized (Minimum Presence Time) How quickly can a resource be released (Release Time) Measures can be used to characterize the degree of agility of a system Advanced Decision Architectures Collaborative Technology Alliance Components and Context Advanced Decision Architectures Collaborative Technology Alliance Mockets Communications Library Overview Application-level Communications library Replaces TCP and UDP Available for C++, Java, and C# Supports Stream and Message abstractions, with multiple types of service Integrates with KAoS for Bandwidth Control Benefits Better performance than TCP Enhanced API provides new capabilities Message Tagging Replacement Prioritization Detailed statistics and feedback support application adaptation Easy to integrate into existing applications that use sockets Flexible allows easy modifications and customizations Bandwidth control Endpoint migration and terminal mobility Advanced Decision Architectures Collaborative Technology Alliance Mockets in the Field Horizontal Fusion Ft. Benning (2003 & 2004) C4ISR On The Move Ft. Dix ( ) AFRL / NGC J-STARS Flight Test (2007 / 2008) Empire Challenge China Lake (2009) Advanced Decision Architectures Collaborative Technology Alliance Mockets Performance (1) WiFi-based Ad-Hoc Networks PSC-5 Tactical Radio Link Advanced Decision Architectures Collaborative Technology Alliance Mockets Performance (2) Timeliness of Arrival of Data with Message Replacement (Blue Force Tracking Application) Average Bandwidth Used for Communication 20 KB/sec Snapshot when Bandwidth Drops to 5 KB/sec Average Mockets Latency 50 ms significantly faster than other protocols Advanced Decision Architectures Collaborative Technology Alliance Group Manager Group Manager Supports Node and Resource Discovery Designed to Fit Well in MANET Environments Supports Managed and Peer Groups, with Optional Security Mechanisms Flexible in Allowing Nodes to Advertise Weakly or Strongly Flexible in Allowing Arbitrary Propagation of Information Supports combination of both proactive and reactive modes Proactive mode push information out to specified radius Reactive mode search information within specified radius Supports Peer-to-Peer Search Using a Modified Gnutella-style Algorithm Advanced Decision Architectures Collaborative Technology Alliance Group Manager in the Field C4ISR On The Move Ft. Dix ( ) Empire Challenge China Lake (2009) Advanced Decision Architectures Collaborative Technology Alliance Group Manager Results (1) Compare Service Manager / Group Manager with JXTA Measure bandwidth utilized for service discovery operation Use a MANET configuration with intermittent connectivity Target environment for Agile Computing, SOSCOE JXTA designed for the Internet not well suited to networks with a high churn rate For example assumes some stability - uses a Distributed Hash Table But JXTA is still proposed for discovery in such environments Scenario Network with Three - Ten nodes Two to Four UAVs One to Six Ground Nodes Advanced Decision Architectures Collaborative Technology Alliance Group Manager Results (2) Bandwidth Utilization Trend Group Manager scales better Still need to try with more (>20) nodes ARL is currently experimenting with Group Manager and SOSCOE Advanced Decision Architectures Collaborative Technology Alliance AgServe Service-oriented architecture for tactical environments COTS SOAs not well suited Support dynamic services Definition, instantiation, invocation, and relocation Monitor service resource utilization and service invocation patterns Learn resource profiles and invocation patterns Resources include CPU, memory, storage, communications Dynamically manage service instances and invocation mappings Advanced Decision Architectures Collaborative Technology Alliance Service Migration Results Recall - Agility may be measured using Exploit Time Minimum Presence Time Release Time Two experiments with service migration measure agility of AgServe Advanced Decision Architectures Collaborative Technology Alliance Service Migration Experiment 1 Experiment 1 - Determine first two agility measures Time to take advantage of a new resource Minimum time necessary to gain benefit Advanced Decision Architectures Collaborative Technology Alliance Service migration starts providing a benefit (reduction in overall execution time) when transient node is available for > 8 seconds Need to measure with additional services Service Migration Results 1 Advanced Decision Architectures Collaborative Technology Alliance Service Migration Experiment 2 Experiment 2 Release Time Measure Advanced Decision Architectures Collaborative Technology Alliance Agility Measures Summary Initial results promising Still need additional experimentation Advanced Decision Architectures Collaborative Technology Alliance FlexFeed Publish-Subscribe Distribute Processing along the data path Optimize Bandwidth Construct Deterministic and Adaptive Data Distribution Paths Sensor Client A Client B Client C idle Sensor Client A Client B Client C idle Sensor Client A Client B Client C Svc Efficient Data Distribution Advanced Decision Architectures Collaborative Technology Alliance FlexFeed FlexFeed Generates and Manages Data Flow Graphs Specifies data sources, processing elements, and data sinks Also specifies bandwidth and processing requirements Data Flow Graphs Are Mapped Dynamically to the Environment Processing elements are allocated to intermediate nodes based on communications and resource availability Multiple Data Flow Graphs Need to be Concurrently Supported Sensor DB P.E. Sensor P.E. User Advanced Decision Architectures Collaborative Technology Alliance DisService Tactical Data Dissemination Combat Situations Require Timely and Efficient Dissemination of Data Dissemination should be peer-to-peer Well suited for mobile ad-hoc networks Three broad categories Situation Awareness (SA) Data widely disseminated Directed Data usually sent to a small subset of nodes On-demand Data only sent upon request Dissemination Capability Should Support all Three Modes Advanced Decision Architectures Collaborative Technology Alliance Scenario Sensor Data Harvesting Collecting and Disseminating Data from Sensor Networks Advanced Decision Architectures Collaborative Technology Alliance ARL Sensors and Radios IR and Visual Camera Sensors Tripwire sensor with Integrated Visual Camera and Ad-Hoc Radio Ad-Hoc Radio Node Advanced Decision Architectures Collaborative Technology Alliance DisService Results (1) SA Data Scenario Advanced Decision Architectures Collaborative Technology Alliance DisService Results (2) Data Harvesting Scenario Advanced Decision Architectures Collaborative Technology Alliance Conclusions Significant Progress and Results Made Possible By Close collaboration with ARL researchers Participating in the field Freedom to start from the scratch not constrained by existing technologies, COTS, GOTS Research Results Introduction and definition of notion of Agile Computing Significant performance improvements over current state-of-the-art Effective transition to ARL and utilization in numerous field demonstrations / exercises Next Steps Further Integration with ARL radio platforms (e.g., CSR) Further development, experimentation, and transition of DisService Documentation / Distribution / Publications / Transitions