Overview of the OMG Data Distribution Service Douglas C. Schmidt & Jeff Parsons {schmidt,parsons}@dre.vanderbilt.edu http://www.dre.vanderbilt.edu/~schmidt/ Professor of EECS Vanderbilt University Nashville, Tennessee

Past R&D Successes: Platform-centric Systems From this design paradigm… Air Frame Nav WTS Legacy DoD systems are designed to be: Stovepiped Proprietary Brittle & non-adaptive Expensive to develop & evolve Vulnerable AP FLIR GPS IFF Cyclic Exec Problem: Small changes can break nearly anything & everything

Past R&D Successes: Platform-centric Systems …and this operation paradigm… Real-time QoS requirements for platform-centric DoD systems: Ensure end-to-end QoS, e.g., Minimize latency, jitter, & footprint Bound priority inversions Allocate & manage resources statically Utility Resources Utility “Curve” “Broken” “Works” “Harder” Requirements Problem: Lack of any resource can break nearly anything & everything

Past R&D Successes: Network-centric Systems …to this design paradigm… Today’s leading-edge DoD systems are designed to be: Layered & componentized More standard & COTS Robust to expected failures & adaptive for non-critical tasks Expensive to evolve & retarget Vulnerable Air Frame AP Nav WTS Event Channel Replication Service GPS IFF FLIR Object Request Broker Problem: Unanticipated changes can break many things

Past R&D Successes: Network-centric Systems …and this operational paradigm… Applications Applications Interceptor Sys Cond Sys Cond Sys Cond Sys Cond Interceptor Middleware } Mechanism & Property Managers { Workload & Replicas Workload & Replicas Local Resource Managers Connections & priority bands Connections & priority bands Local Resource Managers QoS Contracts QoS Contracts CPU & memory CPU & memory Network latency & bandwidth Network latency & bandwidth Endsystem Endsystem Adaptive Real-time Middleware

Past R&D Successes: Network-centric Systems …and this operational paradigm… Problem: Network-centricity is an afterthought in today’s systems Resources Utility Desired Curve “Working Range” “Softer” Requirements

New Demands on Enterprise DRE Systems Mapping & integrating problem artifacts to solution artifacts is very hard Key challenges in the problem space Network-centric, dynamic, very large-scale “systems of systems” Stringent simultaneous quality of service (QoS) demands Highly diverse & complex problem domains Key challenges in the solution space Enormous accidental & inherent complexities Continuous evolution & change Highly heterogeneous platform, language, & tool environments

DARPA PCES Capstone demo, 4/14/05, White Sands Missile Range Case Study: QoS-enabled Publish/Subscribe Technologies for Tactical Information Management Feedback & Control Coordination Of Multiple UAVs Dynamic Mission Replanning Image Processing & Tracking DARPA PCES Capstone demo, 4/14/05, White Sands Missile Range

Case Study: QoS-enabled Publish/Subscribe Technologies for Tactical Information Management Feedback & Control Coordination Of Multiple UAVs Dynamic Mission Replanning Image Processing & Tracking Synchronization Memory Management Physical Access Asynchronous Event Handling Scheduling Transfer of Control Aspect Languages Modeling Tools Real-time JVMs Model Checking Real-time ORBs

Real-time Event Service Limitations with Demo’d PCES Information Management Technologies Shooter information management was based on platform-centric Real-time CORBA & Real-time CORBA Event Service Well-suited for point-to-point & static pub/sub command processing over wireline networks e.g., statically provisioned QoS policies Poorly suited for dynamic pub/sub info dissemination to multiple nodes in MANETs e.g., too many layers, excessive time/space overhead, inflexible QoS policies & pub/sub model, etc. Real-time Event Service Tactical Network & RTOS Object Request Broker Real-time Info to Cockpit Problem: Net-centricity is afterthought in platform-centric technologies

Problem: Enterprise technologies are ill suited for tactical systems Limitations with Demo’d PCES Information Management Technologies Track Processing C2 & C4ISR information management based on J2EE & Java Messaging Service (JMS) Best suited for operational enterprise environments e.g., large data centers connect via wireline networks Poorly suited for tactical environments e.g., lack of QoS policies & RTOS integration, extremely high time/space overhead Java Messaging Service J2EE Middleware Enterprise Network & OS Problem: Enterprise technologies are ill suited for tactical systems

Goal is “better than best-effort,” subject to resource constraints… Our R&D Goal: Evaluate QoS-enabled Info Brokering for Tactical Systems Solutions must function properly where Communication bandwidth is limited/variable Connectivity is intermittent Connections are noisy Processing & storage capacity are limited Power & weight limits affect usage patterns Unanticipated workflows are common Dynamic network topology & membership changes are frequent Ideally, solutions should be COTS, standard, & integrate with heterogeneous legacy systems Goal is “better than best-effort,” subject to resource constraints…

RT Info to Cockpit & Track Processing Overview of the Data Distribution Service (DDS) DDS is an highly efficient OMG pub/sub standard e.g., fewer layers, less overhead Topic R Data Writer R Data Reader R Publisher Subscriber RT Info to Cockpit & Track Processing DDS Pub/Sub Infrastructure Tactical Network & RTOS

Overview of the Data Distribution Service (DDS) DDS is an highly efficient OMG pub/sub standard e.g., fewer layers, less overhead DDS provides meta-events for detecting dynamic changes Topic R NEW TOPIC Data Writer R Data Reader R NEW SUBSCRIBER Publisher Subscriber NEW PUBLISHER

Overview of the Data Distribution Service (DDS) DDS is an highly efficient OMG pub/sub standard e.g., fewer layers, less overhead DDS provides meta-events for detecting dynamic changes DDS provides policies for specifying many QoS requirements of tactical information management systems, e.g., Establish contracts that precisely specify a wide variety of QoS policies at multiple system layers HISTORY Topic R RESOURCE LIMITS Data Writer R S1 Data Reader R S2 S3 S4 S5 Publisher S6 Subscriber S7 LATENCY X S7 S7 S6 S5 S4 S3 S2 S1 COHERENCY RELIABILITY

Overview of the Data Distribution Service (DDS) DDS is an highly efficient OMG pub/sub standard e.g., fewer layers, less overhead DDS provides meta-events for detecting dynamic changes DDS provides policies for specifying many QoS requirements of tactical information management systems, e.g., Establish contracts that precisely specify a wide variety of QoS policies at multiple system layers Move processing closer to data DESTINATION FILTER Topic R Data Writer R S1 Data Reader R SOURCE FILTER S2 S3 S4 S5 Publisher S6 Subscriber S7 TIME-BASED FILTER

Promising Approach: The OMG Data Distribution Service (DDS) Application read Application write write ‘Global’ Data Store Application write write Application read read Application Workload & Replicas Connections & priority bands CPU & memory Network latency & bandwidth DDS provides flexibility, power, & modular structure by decoupling: Location – anonymous pub/sub Redundancy – any number of readers & writers QoS – async, disconnected, time-sensitive, scalable, & reliable data distribution at multiple layers Platform & protocols – portable & interoperable

DDS Architectural Elements Data-Centric Publish-Subscribe (DCPS) The lower layer APIs apps can use to exchange topic data with other DDS-enabled apps according to designated QoS policies Data Local Reconstruction Layer (DLRL) The upper layer APIs that define how to build a local object cache so apps can access topic data as if it were local DDS spec only defines policies & interfaces between application & service Doesn’t address protocols & techniques for different actors implementing the service Doesn’t address management of internal DDS resources

DDS Application Architecture { The Application Application Application Application Application DLRL DLRL DLRL DLRL DCPS Communication

DDS Domains & Domain Participants The domain is the basic construct used to bind individual applications together for communication Like a VPN Domain 2 Domain 3 2 Domain 1 3 1 1 Node Node Node 2 3 1 1 Node Node Node DomainParticipant

DCPS Entities DCPS Entities include Data can be accessed in two ways Topics Typed data Publishers Contain DataWriters Subscribers Contain DataReaders DomainParticipants Entry points Data can be accessed in two ways Wait-based (synchronous calls) Listener-based (asynchronous callbacks) Sophisticated support for filtering e.g., Topic, Content-FilteredTopic, or MultiTopic Configurable via (many) QoS policies Topic Topic Topic Domain Participant Data Reader Data Writer Data Writer Data Reader Data Reader Data Writer Subscriber Publisher Subscriber Publisher Data Domain

QoS Policies Supported by DDS DCPS entities (i.e., topics, data readers/writers) configurable via QoS policies QoS tailored to data distribution in tactical information systems Request/offered compatibility checked by DDS DEADLINE Establishes contract regarding rate at which periodic data is refreshed LATENCY_BUDGET Establishes guidelines for acceptable end-to-end delays TIME_BASED_FILTER Mediates exchanges between slow consumers & fast producers RESOURCE_LIMITS Controls resources utilized by service RELIABILITY (BEST_EFFORT, RELIABLE) Enables use of real-time transports for data HISTORY (KEEP_LAST, KEEP_ALL) Controls which (of multiple) data values are delivered DURABILITY (VOLATILE, TRANSIENT, PERSISTENT) Determines if data outlives time when they are written … and many more …

Best Effort Reliability QoS in DDS QoS Reliability = BEST_EFFORT Subscriber Subscriber Publisher Subscriber Notification of new data objects timeout deadline time-based filter no notification notification time Very predictable Data is sent without reply Publishers and subscribers match and obey QoS Deadline policy settings Time-based Filter QoS policy gives bandwidth control

Reliable QoS in DDS QoS Reliability = RELIABLE Topic R history Data Writer R S1 Data Reader R S2 S3 S4 S5 Publisher S6 Subscriber S7 S7 S6 S5 S4 S3 S2 S1 Ordered instance delivery is guaranteed

Tradeoff Between Reliability & Determinism QoS Reliability = BEST_EFFORT Can’t have reliability and determinism. Best effort mode for streaming data. No retries of dropped packets. Minimizes latency. Reliable mode for commands & events. Retry dropped packets until timeout. Every packet received in order. Speculative cache mechanism. DDS reliability is tunable. Can be adjusted per subscription. X QoS Reliability = RELIABLE X X

All QoS Policies in DDS Deadline Destination Order Durability Entity Factory Group Data History Latency Budget Lifespan Liveliness Ownership Ownership Strength Partition Presentation Reader Data Lifecycle Reliability Resource Limits Time-Based Filter Topic Data Transport Priority User Data Writer Data Lifecycle

Single vs. Multiple Domain Systems

Data Writers & Publishers Data Writers are the primary access point for an application to publish data into a DDS data domain The Publisher entity is a container to group together individual Data Writers User applications Associate Data Writers with Topics Provide data to Data Writers Data is typically defined using OMG IDL It can therefore be as strongly or weakly typed as you desire

Data Readers & Subscribers A Data Reader is the primary access point for an application to access data that has been received by a Subscriber Subscriber is used to group together Data Readers Subscribers & Data Readers can have their own QoS policies User applications Associate Data Readers with Topics Receive data from Data Readers using Listeners (async) or Wait-Sets (sync)

Types & Keys A DDS Type is represented by a collection of data items. e.g. “IDL struct” in the CORBA PSM struct AnalogSensor { string sensor_id; // key float value; // other sensor data }; A subset of the collection is designated as the Key. The Key can be indicated by IDL annotation in CORBA PSM, e.g., #pragma DDS_KEY AnalogSensor::sensor_id It’s manipulated by means of automatically-generated typed interfaces. IDL compiler may be used in CORBA PSM implementation A Type is associated with generated code: AnalogSensorDataWriter // write values AnalogSensorDataReader // read values AnalogSensorType // can register itself with Domain

Topics A DDS Topic is the connection between publishers & subscribers. A Topic is comprised of a Name and a Type. Name must be unique in the Domain. Many Topics can have the same Type. Provision is made for content-based subscriptions. MultiTopics correspond to SQL join Content-Filtered Topics correspond to SQL select. Domain ID 35 Topic name “MySensor” Type name “AnalogSensor” string sensor_id [ key ] float value

Topic-Based Publish/Subscribe Pressure Temperature DataWriter is bound (at creation time) to a single Topic DataReader is bound (at creation time) with one or more topics (Topic, ContentFilteredTopic, or MultiTopic) ContentFilteredTopic & MultiTopic provide means for content-based subscriptions & “joins”, respectively

Subscription = Topic + DataReader application Topic 2 Topic n Data Reader subscriber QoS

Content-based Subscriptions ContentFilteredTopic (like a DB-selection) Enables subscriber to only receive data-updates whose value verifies a condition. e.g. subscribe to “Pressure” of type AnalogData where “value > 200” MultiTopic (like a DB-join operation) Enables subscription to multiple topics & re-arrangement of the data-format e.g. combine subscription to “Pressure” & “Temperature” & re-arrange the data into a new type: struct { float pres; float temp; };

DDS Content-Filtered Topics Topic Instances in Domain Instance 1 Value = 249 Instance 2 Value = 230 Content-Filtered Topic Instance 3 Value = 275 Topic Instance 4 Value = 262 Filter Expression: Value > 260 Instance 5 Value = 258 Instance 6 Value = 261 Expression Params Instance 7 Value = 259 Filter Expression and Expression Params determine which Topic instances the Subscriber receives.

DDS MultiTopic Subscriptions Domain Participant Domain Participant Data Reader Data Reader Data Reader Data Reader Subscriber Subscriber Subscriber MultiTopics can combine, filter, and rearrange data from multiple Topics.

Example: Create Domain Participant DomainParticipant object acts as factory for Publisher, Subscriber, Topic and MultiTopic entity objects // used to identify the participant DomainId_t domain_id = ...; // get the singleton factory instance DomainParticipantFactory_var dpf = DomainParticipantFactory::get_instance (); // create domain participant from factory DomainParticipant_var dp = dpf->create_participant (domain_id, PARTICIPANT_QOS_DEFAULT, NULL);

Example: Create Topic FooDataType foo_dt; foo_dt.register_type (dp, …… // register the data type associated with the topic FooDataType foo_dt; foo_dt.register_type (dp, “Foo”); // create a topic Topic_var foo_topic = dp->create_topic (“Foo_topic”, //topic name “Foo”, // type name TOPIC_QOS_DEFAULT, // Qos policy NULL); // topic listener

Example: Create Subscriber and DataReader …… // create a subscriber from the domain participant SubscriberQos sQos; dp->get_default_subscriber_qos (sQos); Subscriber_var s = dp->create_subscriber (sQos, NULL); // create a data reader from the subscriber // and associate it with the created topic DataReader_var reader = s->create_datareader (foo_topic.in (), DATAREADER_QOS_DEFAULT, FooDataReader_var foo_reader = FooDataReader::_narrow (reader.in ());

Example: Create Publisher and DataWriter …… // create a publisher from the domain participant PublisherQos pQos; dp->get_default_publisher_qos (pQos); Publisher_var p = dp->create_publisher (pQos, NULL); // create a data writer from the publisher // and associate it with the created topic DataWriter_var writer = p->create_datawriter (foo_topic.in (), DATAWRITER_QOS_DEFAULT, NULL); // narrow down to specific data writer FooDataWriter_var foo_writer = FooDataWriter::_narrow (writer); // publish user-defined data Foo foo_data; foo_writer->write (foo_data);

How to Get Data (Async Listener-based) Listener_var subscriber_listener = new MyListener(); foo_reader->set_listener(subscriber_listener); MyListener::on_data_available(DataReader reader) { FooSeq_var received_data; SampleInfoSeq_var sample_info; reader->take(received_data.out (), sample_info.out (), ANY_SAMPLE_STATE, ANY_LIFECYCLE_STATE); // Use received_data …… }

How to Get Data (Sync Wait-based) Condition_var foo_condition = reader->create_readcondition(ANY_SAMPLE_STATE, ANY_LIFECYCLE_STATE); WaitSet waitset; waitset->attach_condition(foo_condition); ConditionSeq_var active_conditions; Duration_t timeout = {3,0}; waitset->wait(active_conditions.out (), timeout); ... FooSeq_var received_data; SampleInfoSeq_var sample_info; reader->take_w_condition(received_data.out (), sample_info.out (), foo_condition); // Use received_data

Benchmark Scenario Two processes perform IPC in which a client initiates a request to transmit a number of bytes to the server along with a seq_num (pubmessage), & the server simply replies with the same seq_num (ackmessage). The invocation is essentially a two-way call, i.e., the client/server waits for the request to be completed. The client & server are collocated. DDS & JMS provides topic-based pub/sub model. Notification Service uses push model. SOAP uses p2p schema-based model.

Testbed Configuration Hostname blade14.isislab.vanderbilt.edu OS version (uname -a) Linux version 2.6.14-1.1637_FC4smp (bhcompile@hs20-bc1-4.build.redhat.com) GCC Version g++ (GCC) 3.2.3 20030502 (Red Hat Linux 3.2.3-47.fc4) CPU info Intel(R) Xeon(TM) CPU 2.80GHz w/ 1GB ram

Test Parameters // Complex Sequence Type struct Inner { string info; long index; }; typedef sequence<Inner> InnerSeq; struct Outer { long length; InnerSeq nested_member; typedef sequence<Outer> ComplexSeq; Average round-trip latency & dispersion Message types: sequence of bytes sequence of complex type Lengths in powers of 2 Ack message of 4 bytes 100 primer iterations 10,000 stats iterations

Roundtrip Latency Results (1/2)

Roundtrip Latency Results (2/2)

Roundtrip Latency Results Analysis From the results we can see that DDS has significantly better performance than other SOA & pub/sub services. Although there is a wide variation in the performance of the DDS implementations, they are all at least twice as fast as other pub/sub services.

Encoding/Decoding Benchmarks Measured overhead and dispersion of encoding C++ data types for transmission decoding C++ data types from transmission DDS3 and GSOAP implementations compared Same data types, platform, compiler and test parameters as for roundtrip latency benchmarks

Encoding/Decoding Results (1/2)

Encoding/Decoding Results (2/2)

Encoding/Decoding Results Analysis Slowest DDS implementation is compared with GSOAP. DDS is faster. Almost always by a factor of 10 or more. GSOAP is encoding XML strings. Difference is larger for byte sequences. DDS implementation has optimization for byte seq. Encodes sequence as a single block – no iteration. GSOAP always iterates to encode sequences. Jitter discontinuities occur at consistent payload sizes.

Summary of DCPS features DDS subscribers publishers Information consumer subscribe to information of interest Information producer publish information DDS matches & routes relevant info to interested subscribers Efficient Publish/Subscribe interfaces QoS suitable for real-time systems deadlines, levels of reliability, latency, resource usage, time-based filter Listener & wait-based data access suits different application styles Support for content-based subscriptions Support for data-ownership Support for history & persistence of data-modifications

Data Local Reconstruction Layer (DLRL) Track Track 3D_Track DLRL Cache Track_Topic 3D -Track DCPS

Goals of DLRL Data Local Reconstruction Layer (DLRL) model is local to an application “Object-oriented” access to data Application developers can describe classes with their methods, data fields, & relations attach some of those data fields to DCPS entities manipulate these objects (i.e., create, read, write, delete) using native language constructs activate attached DCPS entities to update objects have these objects managed in a cache Like a mapping or binding (intuition only)

DLRL Objects DLRL objects are (native) language objects with: data members and methods Only the data members are relevant to data distribution; they can be: attributes, i.e., values relations, i.e., reference other objects plain local data members (i.e., not known or involved in data distribution) are also supported DLRL classes can be organised by inheritance DLRL objects maps to one or more DCPS Topics

DLRL Object Examples

DLRL Radar Example Oid x y Class radar Oid comments Class index Oid z Track x : real y : real comments [*] : string w : integer Track3D z : real Radar tracks a_radar * 0..1 Oid 1 2 x 100 102 y 200 201 TRACK_TOPIC Class Track Track3D radar 11 COMMENTS_TOPIC Oid 1 comments a comment another comment Class Track index 1 Oid 2 z 300 T3D_TOPIC Oid 11 RADAR_TOPIC R_Oid 11 RADAR_TRACKS_TOPIC T_Oid 1 2 T_Class Track Track3D index 1

Evaluating DDS Pros Low overhead & efficient use of transport bandwidth e.g., less features/overhead than CORBA in the main request path Dynamically scalable & highly flexible e.g., can receive notification about all sorts of meta-events, such as new topics, publishers, subscribers, etc. Supports one-to-one, one-to-many, many-to-one, & many-to-many communication Large number of configuration parameters & QoS policies that give developers extensive control of message transmission & reception Cons DDS is not well suited to request-reply services, file transfer, & transaction processing The data-distribution paradigm is not optimized for sending a request & waiting for a reply Implementations don’t yet cover the entire spec Lack of interoperability between different vendor’s implementations

Comparing CORBA with DDS Distributed object • Client/server • Remote method calls • Reliable transport Best for • Remote command processing • File transfer • Synchronous transactions Distributed data • Publish/subscribe • Multicast data • Configurable QoS Best for • Quick dissemination to many nodes • Dynamic nets • Flexible delivery requirements DDS & CORBA address different needs Complex systems often need both…