Building Parallel Time-Constrained HLA Federates: A Case Study with the Parsec Parallel Simulation Language Winter Simulation Conference (WSC’98), Washington D.C. Tuesday, December 15th, 1998 C. D. PhamR. L. Bagrodia Department of Computer Science University of California, Los Angeles

Outline Overview of HLA and Parallel simulation The rational behind this work The Parsec Language Architecture of a parallel federate with Parsec Conclusions

The HLA framework Runtime Infrastructure (RTI) Federation Management Declaration Management Object Management Ownership Management Time Management Data Distribution Management The High Level Architecture calls for a federation of simulations to achieve interoperability and reuse of software. Federation Without HLA, simulations are mostly independents and interoperability is not easy. Logical simulations Hardware, human-in-the loop real-time simulators Display, statistics Rules for the federation and the federates behavior An Object Model Template to describe the simulation objects An Interface Specification simulator real-time players tools

Real system Objects can have attributes HLA: How to interoperate? Interoperability is achieved by subscription and publication of object’s attributes between the federates in the federation execution. communication node throughput, position radio transmitter power, frequency wired node # links

RTI Ambassador Federate RTI HLA: Publication and Subscription Publication of an attribute means a federate can produce values for that attribute. Federate Ambassador RTI Ambassador I can provide the position for F14 planes publishObjectClass I want to know the position of all F14 planes. startUpdates subscribeObjectClass Attribute reflectAttributesValues Subscription to an attribute means a federate wishes to receive values for that attribute. updateAttributeValues Federate Ambassador

HLA: Time management HLA includes advanced time management services with receive order and time stamped messages FEDERATE Time-constrained 25 RTI RO23RO25 Time-regulated FEDERATE 25 Lookahead = 2 RTI NO! RO Compute a Lower Bound Time Stamp (LBTS) Ask for time advance grant

Parallel Simulation Concepts A model consists of simulation objects (logical processes) and events. LPs are distributed on several processors. Only time stamped events are exchanged between LPs. Synchronization and causality constraints are handled by a conservative or optimistic simulation algorithm.

The rational behind this work Parallel simulation can potentially reduce the simulation time of large scale systems. HLA can benefit from parallel federates in a federation execution to achieve speed-up for the entire federation. Federation sequential parallel real-time players parallel computer NOWs...

Parallel simulation and HLA As fast as possible execution. Conservative or optimistic synchronization among LPs. PADS HLA Mainly real time simulation. Conservative among all the federates. Time management Interactions Event paradigm. In most cases, only time stamped events. Subscription/publication. Received Order and Time Stamped Order.

Parallel federates... are time-constrained and time-regulated. must take into account TSO messages from the RTI, have to report a simulation time, have to ask for time advance grants. contain several logical processes. have to compute a global simulation time, provide non-blocking scheduling regarding time advance grants.

One or several federates? Having several federates sometimes facilitates the design of the model but introduces extra dependencies in a conservative federates It is better to have one federate for the entire set of Logical Processes

Need for an accurate GST Even a conservative federate has to implement a GST computation. The GST is the only simulation time value the RTI knows about the parallel federate. The GST must be computed frequently for the other federate to advance in time.

When and how to advance the time LPs cannot request time advance grants directly from the RTI. Messages from the RTI gain to be directly handled by the synchronization algorithm. It is possible to use the LBTS from the RTI

The PARSEC language LP are programmed as an actor or entity all activities of an entity are initiated on message receipts An LP cannot directly modify the state of another LP Event are represented by message communications e(t,p,a): send message m(e) at time t to LP p On receiving m(e), p executes actions a change its state represented by local variables schedule events (messages) at time >=t LPs are scheduled with a large variety of algorithms Conservative with null-messages, conditional events, synchronous, optimistic and adaptive algorithms. Message-based, process-interaction approach Parallel conservative federates Shared memory architecture RTI v1.0 for Solaris 2.5 ENVIRONMENT

Translation of notifications Notifications from the RTI are translated into a PARSEC message and sent back to the entities. RTI callback reflectAttributeValues c1.a1 h Receive (HLACallback m) { switch (m.type) { case reflect: Update(m.C, m.A); break; case discover: … } has subscribed to c1.a1 Entities can register to receive a given set of notification types. has registered to reflect and discover HLA_registerEntity( self, reflect | discover); Ensure that both TSO and RO messages can be mixed into PARSEC Ensure that the entities receive the messages at the correct time

Time management support Sequential federates RTI Entities interact directly with the RTI HLA_timeAdvanceGrantRequest() Request and wait for a time advance grant using nextEventRequest. ENTITY

Parallel federates Entities interact with the RTI via a time manager entity HLA_addEntitySimclock()HLA_provideEntitySimclock() Indicates to the PARSEC runtime that an entity wishes to participate in the GST computation. Allocates data structures for the calling entity. Provides a simulation time value to the time manager. Performs the GST computation and asks for the time advance request from the RTI. RTI ENTITY

Overview of time management Parallel federates

The wireless model example 1 sequential federate: a weather federate that publishes a weather_condition attribute. 1 parallel federate: a wireless communication network that subscribes to the weather_condition attribute. The transmission quality depends on the weather_condition attribute. Can the parallel federate provide speedup for the entire federation execution?

Conclusions