Next: Aerospace Engineering
Up: Vignettes in HPCC
Previous: Vignettes in HPCC
(Based on information supplied by Frank Blitzer, Honeywell)
Real-Time embedded systems applications are:
- processing intensive - already requiring teraFLOPS ;
- software intensive - with typically 100,000's of lines of code per application;
- dual use - both military and commercial markets (medical systems for example);
The market size will be around $300M per year by the year 2000 for
military applications at least. Deployed systems need highly mobile
processing equipment for both ground-based and airborne applications.
This implies very high reliability of the hardware.
Some typical applications include systems for:
- wide area surveillance;
- command and control;
- battle management;
- real-time mission planning;
- real-time information to the warrior;
- gas and oil exploration;
- ocean floor mapping;
- autonomic ship;
- medical imaging.
The current proposed software development model for real-time software systems
has the following characteristsics:
- algorithm development cycle involves use of distributed work stations
with subsequent port to target (MPP) machine type;
- high algorithm coupling efficiency - between 50% and 75%;
- deployed resources must meet meet real-time schedule constraints;
- balanced resource loading problems occur - both static and dynamic;
- light weight kernel operating systems are needed.
Current barriers to successful development include:
- algorithm mapping performance hindered by poor performance of
workstations in real-time;
- necessary software tools may not be applicable to both
MPP systems and distributed workstations;
- static and dynamic load balancing may not be possible for HPC use
until ported to target system.
The potential result of these barriers is that:
- software development costs will rise;
- will need two sets of tools;
- software sizeing estimates will be incorrect;
- overall cost and schedule will have a negative impact on the user.
The embedded systems tools market includes the real time embedded
military systems market which is expected to be at least $300M per year by
2000, as well as the sizeable non-embedded medical imaging and
diagnostics market. Some sought after characteristics are:
- accurate size estimates of code and HPC ;
- efficient mapping of complex algorithms ;
- capability of providing load-balancing;
- capability of efficient development of millions of lines of code;
- capability of software re-engineering and re-use;
The key challenges are to drive the market with hardware and software to
make a self-sustaining industry and also to use government funding to
prime the market development.
Next: Aerospace Engineering
Up: Vignettes in HPCC
Previous: Vignettes in HPCC
Geoffrey Fox, Northeast Parallel Architectures Center at Syracuse University, gcf@npac.syr.edu