Heterogeneous Systems Modelling with Adaptive Traffic Profiles and Its Application to Worst-Case Analysis of a DRAM Controller

Abstract

Computing Systems are evolving towards more complex, heterogeneous systems where multiple computing cores and accelerators on the same system concur to improve computing resources utilization, resources re-use and the efficiency of data sharing across workloads. Such complex systems require equally complex tools and models to design and engineer them so that their use-case requirements can be satisfied. Adaptive Traffic Profiles (ATP) introduce a fast prototyping technology, which allows one to model the dynamic memory behavior of computer system devices when executing their workloads. ATP defines a standard file format and comes with an open source transaction generator engine written in C++. Both ATP files and the engine are portable and pluggable to different host platforms, to allow workloads to be assessed with various models at different levels of abstraction. We present here the ATP technology developed at Arm and published in [5]. We present a case-study involving the usage of ATP, namely the analysis of the worst-case latency at a DRAM controller, which is assessed via two separate toolchains, both using traffic modelling encoded in ATP.

The Road towards Predictable Automotive High - Performance Platforms

“…A First-Ready, First-Come-First-Served (FR-FCFS) DRAM controller is chosen as an example to show how to compute a service curve for a read request arriving at the latter. To do so, first an upper bound on its WCD [14] is computed. Furthermore, it is shown that this upper bound is not overly pessimistic by computing a lower bound too and that the gap between the upper and lower bound is null to negligible in practical cases.…”

Computationally Efficient Worst-Case Analysis of Flow-Controlled Networks With Network Calculus

“…The latter is a theory for deterministic network evaluation, which dates back to the early 1990s, and it is mainly due to the work of Cruz [7], [8], Le Boudec and Thiran [9], and Chang [10]. Originally devised for the Internet, where it was used to engineer models of service [11]- [15], it has found applications in several other contexts, from sensor networks [16] to avionic networks [17], [18], industrial networks [19]- [21], automotive systems [22] and systems architecture [23], [24]. Its main strength is that it allows one to compute worst-case delay bounds in systems with multi-hop traversal.…”

Extending the Network Calculus Algorithmic Toolbox for Ultimately Pseudo-Periodic Functions: Pseudo-Inverse and Composition

“…NC dates back to the early 1990s, and it is mainly due to the work of Cruz [1,2], Le Boudec and Thiran [3], and Chang [4]. Originally devised for the Internet, where it was used to engineer models of service [5][6][7][8][9], it has found applications in several other contexts, from sensor networks [10] to avionic networks [11,12], industrial networks [13][14][15] automotive systems [16] and systems architecture [17,18].…”

“…Extending the NC Algorithmic Toolbox for UPP Functions bits Fig. 1: Example of leaky-bucket shaper, taken from [17]. The traffic process A(t) is always below the arrival curve α(t) and its translations along A(t).…”