[Master] Adapting Hardware Platforms to a Multi-Core Response Time Analysis Framework

Advisors : Claire Maiza and Bruno Ferres

 Context

Providing strong guarantees on the temporal bevahior of critical-systems is crucial in severous domains, such as avionics, aeronautics, automotive or nuclear. In order to do so, techniques have been developed to provide Worst-Case Execution Times (or WCET) of programs used in those critical systems, making it possible to bound the response time of the software on some processors – namely, on mono-core processors.

However, with the raising of multi- and many-core systems, it becomes more difficult to bound the response time of a given program, as it may interfere with other tasks at runtime. Analysis of the Worst Case Response Time (or WCRT) must hence consider not only the WCET of the task, but also the timing penalties that may occur due to the interferences of the other tasks (e.g., cache and bus accesses, that may introduce unplanned delays). Last but not least, the analysis must also consider the hardware systems on which the software is being executed, as several aspects of the hardware can impact both the WCET and the interferences.

In this context, the Multicore Interference Analysis framework (or MIA) has been proposed in Verimag. This framework can be used to compute the WCRT of a given program, for a given hardware platform. Currently, it supports platforms based on standard features (e.g., single-level round-robin arbitration, L1 cache, . . . ), as well as two generations of the MPPA [1] . The MPPA platforms exhibit interesting hardware features for WCRT analysis — namely, using configurable banked memories to limit interferences — which are not always available on other platforms.

 Goals of the Internship

The goal of this internship is to study the impact of adding new hardware platforms to the MIA framework, in the context of the ANR project CAOTIC [2]. In particular, we wish to investigate how other complex platforms can be adapted to the banked memory model, to ease the modeling of interferences, as well as the development of critical applications.

In consequence, the following contributions are expected :

  • consider how those particular components might be configured to be used in critical-systems ;
  • propose new components from the literature, that would be interesting to analyze in the context of multi-/many-core based critical-systems ;
  • study how to model such components in the existing framework, adapting their behavior to the analysis. A particular focus will be put on communication and memory modeling, as they are the main sources of interferences in multi-/many-core systems.

Contributions are hence expected both on the theoretical aspects of the framework, and on the practical side, ideally leading to the implementation of new hardware platforms in the analysis framework.

 Ideal Applicant

This internship proposal is for Computer Science students at M2 level/last year of engineering school [3]. The applicant must be proficient in the following knowledge/skills :

  • micro-architecture of processors (e.g., pipeline, cache policies, bus arbitration, . . . )
  • ability to read and understand research papers as well as technical documentation
  • interest for verification methods (architecture modeling, WCET computation, . . . )

As the MIA framework is written in python, good programming skills in this language would be appreciated.

 Applications

To apply, send an email to claire.maiza@univ-grenoble-alpes.fr and bruno.ferres@univ-grenoble-alpes.fr, with your resume, a short covering letter, as well as any document that may support your application.

See the proposal attached.

[1Massively Parallel Processor Array, a many-core processor from the Kalray company http://www.kalrayinc.com.

[2Collaborative Actions on Timing Interferences, a research project funded by the Agence Nationale de la Recherche, led by Verimag.

[3Motivated applications at M1 level will also be considered.


Documents joints

proposal_Verimag

22 octobre 2024
info document : PDF
134.1 kio

Contact | Plan du site | Site réalisé avec SPIP 4.2.16 + AHUNTSIC [CC License]

info visites 4187186