Analysis and optimization of mixed-criticality applications on partitioned distributed architectures

Abstract

In this paper we are interested in mixed-criticality applications implemented using distributed heterogenous architectures, composed of processing elements (PEs) interconnected using the TTEthernet protocol. At the PE-level, we use partitioning, such that each application is allowed to run only within predefined time slots, allocated on each processor. At the communication-level, TTEthernet uses the concepts of virtual links for the separation of mixed-criticality messages. TTEthernet integrates three types of traffic: Time-Triggered (TT) messages, transmitted based on schedule tables, Rate Constrained (RC) messages, transmitted if there are no TT messages, and Best Effort (BE) messages. We assume that applications are scheduled using Static Cyclic Scheduling (SCS) or Fixed-Priority Preemptive Scheduling (FPS). We are interested in analysis and optimization methods and tools, which decide the mapping of tasks to PEs, the sequence and length of the time partitions on each PE and the schedule tables of the SCS tasks and TT messages, such that the applications are schedulable and the response times of FPS tasks and RC messages is minimized. We have proposed a Tabu Search-based meta-heuristic to solve this optimization problem, which has been evaluated using several benchmarks.

Timing analysis of mixed-criticality hard real-time applications implemented on distributed partitioned architectures

“…This section presents the application model of applications scheduled with FPS. For the full model the reader is referred to [14]. We consider that the mapping of tasks to processing elements (PEs) and the assignment of tasks to partitions is given, and can be determined using our approach from [14].…”

“…We assume that the hardware and software architecture implements a temporal-and space-partitioning scheme similar to IMA [14].…”

“…The mapping and the partitioning are given, and we assume they are derived with our approach from [14]. The hashed partition slices represent partitions on which the application A 1 is not allowed to execute.…”

“…The algorithm starts by initializing the extended completion time of p ab with p ab 's original completion time and the availability and demand with 0 (lines 10-12). It then proceeds to iteratively re-compute the availability and demand until the availability is greater or equal to the demand (lines [13][14][15][16][17][18][19]. At each iteration, the extended completion time of p ab is increased with the difference between the current values of the availability and demand (lines 16-18).…”

“…Researchers have addressed systems with mixed time-criticality requirements, showing how TT/ET tasks [10] can be integrated onto the same platform. Researchers have also started to address the integration of mixed safetycriticality tasks onto the same architecture [2,14]. In [14], we have proposed an optimization approach to determine the mapping of tasks to PEs, the assignment of tasks to partitions, the sequence and size of the time slots on each PE and the schedule tables, such that all the applications are schedulable and the development costs are minimized.…”