Optimization

What specific environments does Gedae support?

Gedae gives the user access to vendor-supported methods of data communication between processors. Different vendors support different methods of data transfer. Gedae communication is built using the services provided by the vendor's operating system or board support software. Gedae does not provide unique communications. Typically, sockets shared memory and DMA Communication Mechanisms are provided.

Gedae provides the user with the ability to select from available supported types of communication to be used between specific processors. The type is vendor and system topology specific.

How much does Gedae support RT control systems? For example, weapons control or flight control?

Discussion

Gedae can be used effectively to handle real time performanceconstraints. Real time systems require:

• Ability to keep up with incoming data,
• Guaranteed processing latency,
• Periodic scheduling of services, and
• Ability to respond quickly to interrupts.

A Gedae developer has several tools for meeting such requirements. A Gedae application consists of multiple processors each dynamically scheduling static schedules. Gedae provides control over both the dynamic and the static scheduling of a Gedae graph giving the developer the control he needs to meet real time performance criteria.

A static schedule is a precalculated list of primitive function firings that does the work of the application. Static schedules have the advantage of predictability and repeatability - important considerations in a real time application. They also require very low scheduling overhead as the schedules are determined prior to run time. However some scheduling decisions need to be made at runtime such as when a dynamic box steers data down one branch of a graph or another. Gedae handles major portions of the graph as predictable static schedules and only uses dynamic scheduling where needed to implement dynamic graph behavior.

Solution

The execution of both the static and dynamic schedules can be tuned by the developer to help insure that real time constraints and throughput requirements are met. In addition graphs can be easily parallelized in Gedae. The following shows some of the capabilities a developer has to make real-time performance possible:

1. Partitioning and Mapping: Graphs can be partitioned and mapped to multiple processors to get the parallelism needed to keep up with real-time input data.

2. Static Schedules Granularity Control: The user can control a static schedule by setting overall schedule granularity and the granularity of individual function boxes. This allows the developer to tradeoff latency and throughput.

3. Static Schedule Priority: Frequently in a static schedule several primitives may be able to fire at a given point. Gedae allows the default decision made by the static scheduler to be overridden by user set priorities.

4. Periodic Scheduling: A static schedule can be marked to run periodically (e.g., every 10msec) by the developer. Other timeout and retry options can be set by the developer.

5. Flexible Scheduling: A static schedule can be marked to run flexibly. That is the schedule will adjust granularity on the fly to handle the available data thus reducing latency.

6. Round-Robin: A static schedule can be marked to run round-robin insuring fairness.

7. Dynamic Scheduling Priorities: The priority of a static schedule can be set determining which of the runnable static schedules will fire first.

8. Interrupt Handling: Interrupt handlers can be incorporated into Gedae graphs.

More Information