Conference Paper

LIFO-Backpressure achieves near optimal utility-delay tradeoff

Dept. of Electr. Eng., Univ. of Southern California, Los Angeles, CA, USA
DOI: 10.1109/WIOPT.2011.5930067 Conference: Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks (WiOpt), 2011 International Symposium on
Source: IEEE Xplore

ABSTRACT There has been considerable recent work developing a new stochastic network utility maximization framework using Backpressure algorithms, also known as MaxWeight. A key open problem has been the development of utility-optimal algorithms that are also delay efficient. In this paper, we show that the Backpressure algorithm, when combined with the LIFO queueing discipline (called LIFO-Backpressure), is able to achieve a utility that is within O(1/V) of the optimal value for any scalar V ≥ 1, while maintaining an average delay of O([log(V)]2) for all but a tiny fraction of the network traffic. This result holds for general stochastic network optimization problems and general Markovian dynamics. Remarkably, the performance of LIFO-Backpressure can be achieved by simply changing the queueing discipline; it requires no other modifications of the original Backpressure algorithm. We validate the results through empirical measurements from a sensor network testbed, which show good match between theory and practice.

  • Source
    • "V * N ) be a vector of Lagrange multipliers , which solves the dual problem (9) with parameter V . The following theorem from [9] describes a steady state property of the drift-plus-penalty algorithm: "
    [Show abstract] [Hide abstract]
    ABSTRACT: One practical open problem is the development of a distributed algorithm that achieves near-optimal utility using only a finite (and small) buffer size for queues in a stochastic network. This paper studies utility maximization (or cost minimization) in a finite-buffer regime and considers the corresponding delay and reliability (or rate of packet drops) tradeoff. A floating-queue algorithm allows the stochastic network optimization framework to be implemented with finite buffers at the cost of packet drops. Further, the buffer size requirement is significantly smaller than previous works in this area. With a finite buffer size of $B$ packets, the proposed algorithm achieves within $O(e^{-B})$ of the optimal utility while maintaining average per-hop delay of $O(B)$ and an average per-hop drop rate of $O(e^{-B})$ in steady state. From an implementation perspective, the floating-queue algorithm requires little modification of the well-known Drift-Plus-Penalty policy (including MaxWeight and Backpressure policies). As a result, the floating-queue algorithm inherits the distributed and low complexity nature of these policies.
  • Source
    • "Finally, some recent ideas [6] [7] could be incorporated in an orthogonal way to improve analogously the delay performance of all policies, but this is beyond the scope of this paper. Here, we take holistic approaches in designing efficient delay-aware backpressure algorithms, which are both practical – have low computational overhead and are robust to topology changes. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Packet scheduling/routing in wireless ad hoc networks is a fundamental problem for ad hoc networking. Backpressure routing is a solid and throughput optimal policy for such networks, but suffers from increased delays. In this article, we present two holistic approaches to improve upon the delay problems of backpressure - type algorithms. We develop two scheduling algorithms, namely Voting backpressure and Layered backpressure routing, which are throughput optimal. We experimentally compare the proposed algorithms against state-of-the-art delay-aware backpressure algorithms, which provide optimal throughput, for different payloads and network topologies, for both static and mobile networks. The experimental evaluation of the proposed delay reduction algorithms attest their superiority in terms of QoS, robustness, low computational complexity and simplicity.
    09/2014; 14(1):1-16. DOI:10.4108/mca.1.4.e5
  • Source
    • "There have been recent works trying to obtain improved utility-delay tradeoff for stochastic systems. For instance, [16], [17], and [18] propose algorithms that can achieve the [O(), O([log(1//)] 2 )] tradeoff. However, all the above algorithms require a convergence time of Θ(1//). "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we investigate the power of online learning in stochastic network optimization with unknown system statistics {\it a priori}. We are interested in understanding how information and learning can be efficiently incorporated into system control techniques, and what are the fundamental benefits of doing so. We propose two \emph{Online Learning-Aided Control} techniques, $\mathtt{OLAC}$ and $\mathtt{OLAC2}$, that explicitly utilize the past system information in current system control via a learning procedure called \emph{dual learning}. We prove strong performance guarantees of the proposed algorithms: $\mathtt{OLAC}$ and $\mathtt{OLAC2}$ achieve the near-optimal $[O(\epsilon), O([\log(1/\epsilon)]^2)]$ utility-delay tradeoff and $\mathtt{OLAC2}$ possesses an $O(\epsilon^{-2/3})$ convergence time. $\mathtt{OLAC}$ and $\mathtt{OLAC2}$ are probably the first algorithms that simultaneously possess explicit near-optimal delay guarantee and sub-linear convergence time. Simulation results also confirm the superior performance of the proposed algorithms in practice. To the best of our knowledge, our attempt is the first to explicitly incorporate online learning into stochastic network optimization and to demonstrate its power in both theory and practice.
    ACM SIGMETRICS Performance Evaluation Review 04/2014; 42(1). DOI:10.1145/2591971.2591990
Show more


Available from