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

Abstract

In this paper, we address distributed convergence to fair allocations of CPU resources for time-sensitive applications. We propose a novel resource management framework where a centralized objective for fair allocations is decomposed into a pair of performance-driven recursive processes for updating: (a) the allocation of computing bandwidth to the applications (resource adaptation), executed by the resource manager, and (b) the service level of each application (service-level adaptation), executed by each application independently. We provide conditions under which the distributed recursive scheme exhibits convergence to solutions of the centralized objective (i.e., fair allocations). 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. We finally validate our framework with simulations using the TrueTime toolbox in MATLAB/Simulink.

Measurement-based efficient resource allocation with demand-side adjustments

“…In the trivial case of identical tasks, with s 1 = ... = s n , d 1 = ... = d n and λ 1 = ... = λ n , one allocation that satisfies the fairness condition is v * i = 1 /n. A similar fairness measure has been introduced within the context of CPU Bandwidth Allocation problem in [9]. The above definition is more general since the performance indices are not restricted to any specific application scenario and they are only subject to Assumption 3.2.…”

“…The above framework may accommodate resource allocation problems encountered in the context of CPU bandwidth allocation. Recent work [9] has focused on designing such utility functions for the case of time-sensitive applications. In this scenario, the RM is responsible for assigning virtual-platforms v i to each application i.…”

“…Note that the utility function attains a unique maximum when the deadline D i approaches R i , which is the desired property. As described in [9], and in the context of multimedia applications, the soft deadline D i can be considered constant, e.g., D i = h > 0, while the response time can be defined as R i = C i/v i , where C i = θ i s i is the execution time per job (at a service-level s i ), for some θ i > 0, and v i is the speed of execution. In this case, the utility of application i takes on the following form:…”

“…Using Lemma 6.1, we can derive a detailed characterization of the potential attractors of the discrete-time recursions (11), as the following theorem demonstrates. Theorem 6.1 (Global convergence) Consider the overall update recursion (11) with a fixed demand d(k) = d and a step-size satisfying condition (9) and 0 < µ( ) < 1 /γ. Let L denote the limit points 3 of the system of ODE'ṡ…”

“…Motivated by the framework of [9] and the potential of exploiting both measurements available from the tasks and the flexibility of some tasks in changing their operation-level, in this paper we propose a generalized design methodology for addressing a general class of online resource allocation problems. In particular, the RM is responsible for adjusting both the resources and the operationlevels of the task, where the adjustment processes are based only on performance measurements received from the running tasks.…”