Section: New Results
WCET estimation
Participants : Damien Hardy, Benjamin Lesage, Thomas Piquet, Isabelle Puaut.
Predicting the amount of resources required by embedded software is of prime importance for verifying that the system will fulfill its real-time and resource constraints. A particularly important point in hard real-time embedded systems is to predict the Worst-Case Execution Times (WCETs) of tasks, so that it can be proven that task temporal constraints (typically, deadlines) will be met. Our research concerns methods for obtaining automatically upper bounds of the execution times of applications on a given hardware. A particular focus is put on hardware-level analysis (static analysis based on timing models) for multicore platforms.
In 2009, our new results concern the static WCET analysis of tasks running on multicore platforms with shared instruction and/or data caches.
Timing analysis of multicore platforms with shared instruction caches
Participants : Damien Hardy, Thomas Piquet, Isabelle Puaut.
WCET estimation for multicore platforms is a very challenging task because of the possible interferences between cores due to shared hardware resources such as shared caches, memory bus, etc.
We have proposed in [21] a safe WCET estimation method for multi-core architectures with shared non-inclusive instruction cache(s). The method computes the WCET of one task running on one core, in competition with an arbitrary number of other tasks running on the other cores, and thus competing for the shared instruction cache. The proposed method, similarly to [52] , estimates, using static analysis, the worst-case conflicts for the shared caches. However, it is more general than [52] in the sense that it supports multiple levels of shared caches, set-associative caches and an arbitrary number of real-time tasks and cores competing for the shared caches.
While safe, the proposed WCET estimation method might be too pessimistic in case of high pressure for the shared cache level(s). This potential pessimism is reduced in [21] by the proposal of the compiler-directed bypass scheme. Single-usage (not reused) program blocks in shared instruction caches are identified thanks to static analysis of the program code. This enables a compiler-directed bypass scheme, which, from the static knowledge of single-usage blocks, allows a drastic reduction of inter-task and intra-task interferences, and thus a tighter WCET estimate. The experimental results given in [21] , demonstrate the practicality of our approach.
Our ongoing work is to support a larger panel of cache architectures (inclusive caches, exclusive caches).
Evaluation of cache-related migration delays
Participants : Damien Hardy, Isabelle Puaut.
Two approaches may be used for task scheduling in multicore systems: partitioned and global scheduling. Under partitioned scheduling, tasks are assigned to cores and are not allowed to migrate. While this class of approaches imposes no migration overhead, it has the following limitations: task partitioning is a NP-hard problem; dynamic task admittance is hard to support because it would require online re-partitioning. To address these limitations, global scheduling techniques have been proposed. The fundamental premises of these techniques is that task may migrate between cores.
However task migration results in a direct cost, required to save and restore the task context, but also in an indirect cost to reload the cache contents after the migration. After a migration, the reused cache blocks are reloaded in all levels of the cache hierarchy.
Task migration thus results extra cache misses compared with a migration-free execution. Extra cache misses occur in the private cache to load reused blocks as well as on the shared L2 cache when using non inclusive cache hierarchies.
We have proposed in [22] methods to compute safe estimations of the Instruction Cache Related Migration Delay (CRMD) suffered by a task after each migration. Our method relies on static code analysis. Our experimental results demonstrate estimated CRMDs much lower than those that would be obtained with a naive approach.
Timing analysis on multicore platforms with shared data and unified caches
Participants : Damien Hardy, Benjamin Lesage, Isabelle Puaut.
WCET estimation for multicore platforms requires to analyse data cache content as well as instruction cache content. As a first step towards considering the whole memory hierarchy of multicore platforms, we have proposed in [23] a static analysis of data cache hierarchies, based on our previous work published in [5] for instruction cache hierarchies. Our ongoing work concerns the use of (i) compiler-directed bypass to data as well as (ii) cache partitioning methods to decrease or avoid conflicts in shared data or unified caches.
