Distributed management of CPU resources for time-sensitive applications

Abstract

The number of applications sharing the same embedded device is increasing dramatically. Very efficient mechanisms (resource managers) for assigning the CPU time to all demanding applications are needed. Unfortunately existing optimization-based resource managers consume too much resource themselves. In this paper, we address the problem of distributed convergence to efficient CPU allocation for time-sensitive applications. We propose a novel resource management framework where both applications and the resource manager act independently trying to maximize their own performance measure and according to a utility-based adjustment process. Contrary to prior work on centralized optimization schemes, the proposed framework exhibits adaptivity and robustness to changes both in the number and nature of applications, while it assumes minimum information available to both applications and the resource manager. It is shown analytically that efficient resource allocation can be achieved in a distributed fashion through the proposed adjustment process. Experiments using the TrueTime Matlab toolbox show the validity of our proposed approach. AbstractThe number of applications sharing the same embedded device is increasing dramatically. Very efficient mechanisms (resource managers) for assigning the CPU time to all demanding applications are needed. Unfortunately existing optimization-based resource managers consume too much resource themselves.In this paper, we address the problem of distributed convergence to efficient CPU allocation for time-sensitive applications. We propose a novel resource management framework where both applications and the resource manager act independently trying to maximize their own performance measure and according to a utility-based adjustment process. Contrary to prior work on centralized optimization schemes, the proposed framework exhibits adaptivity and robustness to changes both in the number and nature of applications, while it assumes minimum information available to both applications and the resource manager. It is shown analytically that efficient resource allocation can be achieved in a distributed fashion through the proposed adjustment process. Experiments using the TrueTime Matlab toolbox show the validity of our proposed approach.

Introducing service-level awareness in the cloud

“…In the end, the capacity requirements will be roughly proportional to s k i . One of the main differences between this work and similar research in the context of embedded systems [9] is that we do not assume anything about the application's behavior, thus, the RM does not have access to the SL update rules. In fact, our framework is completely general with respect to the choice of the function g i in Eq.…”

Design and implementation of distributed resource management for time-sensitive applications

“…In particular, the proposed scheme: (a) exhibits linear complexity with the number of applications, (b) drops the assumption that the RM has knowledge of application details, and (c) exhibits adaptivity and robustness to the number and nature of applications. This paper extends the theoretical contributions of [6] by addressing global convergence and asynchronous updates. Furthermore, reference [16] presents the full implementation framework in Linux.…”

“…We consider a constant step size > 0, since it provides an adaptive response to changes in the number of applications. In some cases, we will use vector notation, denotingv (6) for the adjustment of resources was motivated by the standard replicator dynamics (cf., [24,Chapter 3]) and in particular the discrete-time equivalent (namely reinforcement learning) introduced in [7]. Note that the RM time complexity is linear with respect to the number of applications, as demonstrated in [16].…”

“…In this scenario, the application i will increase its service level if f i (t k ) > 0, otherwise it will decrease it. Alternative observation terms can also be defined with similar properties as demonstrated in [6].…”

“…If n i=1v i (t k ) ≤ 1, then n i=1v i (t k+1 ) ≤ 1. Thus, for a bounded number of applications, we may pick sufficiently small = (n) such thatv i (t k ) ∈ [0, 1] and n i=1v i (t k ) ≤ 1 for all k = 1, 2, .... We consider now the constrained version of the recursion (6). For some time t k , let us assume that n i=1v i (t k ) ≤ 1, i.e., the allocation is feasible.…”