Principal Investigator

Billed Time – 2 months

Unbilled time – 2 months

Co-Principal Investigators

Objective

Military command and control is becoming increasingly dependent on complex software systems executing on distributed and parallel hardware systems. As humans and computers are integrated into total systems the end-to-end response time of the computer portion of the system becomes a critical factor for effective total systems. It is well established by bitter experiences that ad hoc procedures for estimating the performance of complex applications on parallel and distributed resource environments are not reliable. Failure to attain performance goals has led to costly and highly visible failures in past development programs for complex software systems. Finally, as simulations of behaviors of military systems ranging from nuclear weapons stewardship to virtual reality based combat training migrates to parallel and distributed implementations, the requirement for integration of performance engineering into the development of these systems also becomes important.

The objective for the Performance Oriented End-to-end Modeling System (POEMS) is to meet this requirement for performance engineering of complex, parallel and distributed computer systems to meet operational performance requirements and to enable management of schedule and cost risks from failures to meet performance requirements.

Approach

The POEMS project will attain this objective by creating and experimentally evaluating a problem-solving environment for end-to-end performance modeling of complex parallel/distributed systems, spanning application software, runtime and operating system software, and hardware architecture. The POEMS project combines innovations from communication models, data mediation, parallel programming, performance modeling, software engineering, and CAD/CAE. POEMS is experimentally focused and driven. The Sweep3D wavefront computational kernel has been the first driving application while a kernel from the MSTAR automatic target recognition system will be the second driving application.

The major elements of the conceptual framework of POEMS are: a general model of parallel computation, the use of objects with associative interfaces as the representation basis for components, the use of multiple methods of evaluation of component behavior, incorporation of a knowledge base on the behaviors of common operations on known execution environments and direct derivation of application specifications for computational kernels when they are available.

Performance engineers using POEMS should be able to construct performance models with resolution of detail ranging from early conceptual design to performance tuning of production systems. POEMS will separate specification of applications from specifications for execution environments so that each can be separately varied. POEMS targets supporting performance engineering of a total system at effort levels ranging from 1% to 10% of total system development cost.

Accomplishments for July 1st, 1998 to June 30th 1999

POEMS has, in this year, completed its demonstration that performance models can be composed by integrating analytic, simulation and execution based components. "This demonstration has shown that this integration allows the modeling to be scaled by orders of magnitude while retaining the accuracy of the performance models." POEMS has also established the basis for automation of composition of such performance models. POEMS technology enables incorporation of designing to performance requirements into the mainstream of software system design and development. Many of the specific accomplishments given following are parts of this major achievement.

Demonstration of the effectiveness of POEMS methods through application to studies of the SWEEP3D wave-front code has been accomplished. This demonstration, which included integration of simulation, analytic and execution based evaluation of components, validated the POEMS approach to performance modeling. A significant result of the studies was determination of the scalability properties of the SWEEP3D code on distributed memory parallel architectures.

This demonstration and validation of POEMS was based on development and validation of an initial library of component models (analytic and simulation evaluation based) for the SWEEP3D application and the IBM SP2 and SGI Origin hardware architectures and the MPISIM simulation modeling system for MPI-based communication.

Two major public domain simulators for hardware architectures (RSIM and SimpleScalar) have been integrated into the POEMS component framework. This integration enables incorporation of detailed evaluation of hardware architectures into POEMS modeling studies. SimpleScalar has been configured to model the 604e and the R10000, and these configurations have been validated.

A compiler for automatic synthesis of performance models as static task graphs from applications written in HPF and/or HPF/MPI has been developed. These task graphs can be directly evaluated for performance properties or mapped to the POEMS Specification Language to become components of POEMS models.

Integration of task graph models of applications with the MPI-SIM simulator for communication has demonstrated improvement of factors of 100-1000 in memory usage and 2-5 in simulation time. These improvements enable the simulation of systems or problem sizes that are one to two orders of magnitude larger than was previously possible.

A feasibility demonstration compiler for an initial version of the POEMS Specification Language has been completed.

An initial version of the Performance Knowledge Base has been completed. This version of the PKB includes performance data on commonly used computational algorithms.

 

Current Plan - July 1st 1999 to June 30th, 2000

The POEMS project must, in the next year, complete and automate the compositional process for performance models which has been developed and validated during the first two years of the project and demonstrate these capabilities on a second major application. The current choice for the second major application is the MSTAR automatic target recognition system. The project plan is focused around these goals.

Integration of analytic models of application and operating system components with simulation-based models of processor/memory architectures.

Integration of automatically-generated task graphs with simulation models of

future processors, analytical models of communication system behavior, and

integrated analytical/simulation models of entire systems.

Encapsulation of component models with associative interfaces to enable automation of compositional development of performance models.

Development of a version of the POEMS Specification Language compiler which can support experimental use on a practical system scale.

Completion of component model libraries for: the SWEEP3D and MSTAR applications, the IBM SP2 and SGI Origin hardware architectures and necessary operating system and communication system components.

Testing and validation of the Knowledge Base. Population of the Knowledge Base with POEMS component models. Interfacing of the POEMS Specification Language Compiler with the Knowledge Base.

Mapping of automatically generated task graphs to the POEMS Specification Language. This will enable automation of performance evaluation of current applications on new architectures when the architectures are still in their design phase.

Studies of the effectiveness of evaluation mode integration on the speed-up for evaluation of performance models of large systems and the scale of performance models which can be evaluated.

Execution of a performance modeling study of MSTAR to support parallel and distributed implementations of MSTAR which can accomplish recognition in real-time.

 

Technology Transition

The POEMS project is working directly with the consumers of its technology. The POEMS project includes one of its customers, LANL, as a project participant and the first application,Sweep3D, is a kernel of the ASCI program suite. This application of POEMS to Sweep3D is directly supporting the ASCI program. The current choice for a second application is MSTAR, a DARPA funded project. The MSTAR project office plans a parallel implementation for fiscal 1999 and beyond. POEMS will model MSTAR if the necessary information is available or will otherwise choose another application of direct interest to DARPA.