A Case for Decomposition of FIFO Networks

Ciucu, Florin; Liebeherr, Jörg

doi:10.1109/infcom.2009.5062019

Cited by 3 publications

(11 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is a common assumption in most of the papers on network decomposition [5], [6], [8], [22]. In this paper, we assume that each traffic flow is a Markov-modulated On-Off (MMOO) source which is a common model for voice traffic [18].…”

Section: System Modelmentioning

confidence: 99%

“…Traffic sources which satisfy Eq. (5) and in particular, MMOO sources are also used in the papers studying network decomposition in a non-asymptotic regime [5], [6].…”

Section: System Modelmentioning

confidence: 99%

“…In the literature, two types of convergence are considered: 1-Convergence of D In a FIFO link which is fed by two classes of arrivals, through and cross flows, each consisting of independent, identical leaky-bucket arrivals, the random burstiness of the output of the through flows converges to the input deterministic leakybucket burstiness of the aggregate through flows [22]. Finally, for MMOO traffic flows, the EBB characteristics of the output process in an isolated FIFO queue is shown to converge to those of the input process exponentially fast in N [5], [6]. This provides an even stronger argument for decomposition than the convergence of D u 0 to A 0 .…”

Section: Non-asymptotic Network Decomposition For ∆-Schedulersmentioning

confidence: 99%

“…For a general work-conserving scheduler and a class of Markov-modulated sources [6], it is shown in [5] that O(1) buffers are sufficient to satisfy a target overflow probability. In a multi-flow FIFO scheduler and Markov-modulated On-Off sources, an O( 1 N ) per-flow capacity is shown to be sufficient to guarantee a probabilistic end-to-end delay bound [5].…”

Section: Introductionmentioning

confidence: 99%

“…The viability of network decomposition is considered in asymptotic [8], [17], [20], [22] and numerically in nonasymptotic regimes [5], [6].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Capacity provisioning for schedulers with tiny buffers

Ghiassi-Farrokhfal

Liebeherr²

2013

2013 Proceedings IEEE INFOCOM

View full text Add to dashboard Cite

Abstract-Capacity and buffer sizes are critical design parameters in schedulers which multiplex many flows. Previous studies show that in an asymptotic regime, when the number of traffic flows N goes to infinity, the choice of scheduling algorithm does not have a big impact on performance. We raise the question whether or not the choice of scheduling algorithm impacts the capacity and buffer sizing for moderate values of N (e.g., few hundred). For Markov-modulated On-Off sources and for finite N , we show that the choice of scheduling is influential on (1) buffer overflow probability, (2) capacity provisioning, and (3) the viability of network decomposition in a non-asymptotic regime. This conclusion is drawn based on numerical examples and by a comparison of the scaling properties of different scheduling algorithms. In particular, we show that the per-flow capacity converges to the per-flow long-term average rate of the arrivals with convergence speeds ranging from O log N N to O 1 N depending on the scheduling algorithm. This speed of convergences of the required capacities for different schedulers (to meet a target buffer overflow probability) is perceptible even for moderate values of N in our numerical examples.

show abstract

Section: System Modelmentioning

confidence: 99%

“…Traffic sources which satisfy Eq. (5) and in particular, MMOO sources are also used in the papers studying network decomposition in a non-asymptotic regime [5], [6].…”

Section: System Modelmentioning

confidence: 99%