Computing technologies: platforms, processors, and controllers Chapter | 3 373.3.3 Monitoring and throttling capabilities of real-time flows
An interconnect is in charge of arbitrating accesses to resources that are shared
among the various processing nodes, such as memory or input and output
devices. The parallel requests to shared resources may invoke contentions and
bottlenecks between competing processors and introduce delays to task execu-
tion over multicore or manycore architectures. When designing safety critical
systems in such architectures, it is important to early detect and resolve such
contentions and properly estimate the WCET of tasks. The existing approaches
rely either on hardware or software in order to provide a solution to the problem.
The hardware approach employs a time division multiplexing policy to avoid
conflicts at runtime, whereas the software approaches use specialized execu-
tion models, such as PRedictable Execution Model to separate data access and
computation thus minimizing the risk for conflicts. However, various tasks in
an autonomous system may have different criticality levels, thus it is important
to support mixed-criticality solutions that take advantage of multi or manycore
architectures without adding the conflict risk resolution or prevention overhead
at all time. In such mixed-criticality systems the noncritical tasks receive lower
priority or are interrupted when conflicts reach a threshold level and gain nor-
mal priority back or relaunched when resources are released. In addition to this,
several more flexible strategies that include caching or resource duplication at
the connectivity level have been developed in order to maximize the throughput
of multi or manycore platforms by avoiding conflicts and bottlenecks.
In the case of multicore systems, a hardware contention manager allows
monitoring of the interconnect level activity. The manager improves the sys-
tem’s ability to allocate and de-allocate resources to critical and noncritical
tasks and allows to properly schedule requests to shared resources. The result is
that all critical tasks execute within the provided time-frame at the expense of
maximizing the number of resource requests from noncritical tasks. In the case
of manycore systems, a NoC allows the interconnection of cores. Such chips are
not generic, but rather application-specific, and are designed with characteris-
tics that fit the real-time systems they are serving. The performance evaluation
of a generic NoC in handling resource conflicts between flows is still an open
research topic. A routing mechanism for NoC that will monitor the competing
flows and will facilitate the management of conflicts would also be in favor of
manycore systems. The design of such a mechanism within a NoC interconnect
is a main area of added value for manycore architectures.
3.3.4 Automotive eHPC software environment
Automotive software environments are structured by the AUTOSAR, which is
a partnership between vehicle manufacturers, suppliers of electronics, semicon-
ductors, and hardware and companies from the software industry that operate
on a global level. The AUTOSAR developed software is quite popular among