Showing papers by "Shivendra S. Panwar published in 1997"

A weighted max-min fair rate allocation for available bit rate service

Abstract: An important concept in the ABR service model is the minimum cell rate (MCR) guarantee as well as the peak cell rate (PCR) constraint for each connection. Because of the MCR and PCR constraints, the classical max-min policy no longer suffices to determine the rate allocation since it does not support either the MCR or the PCR. To support the MCR/PCR constraints for each connection, we present the weighted max-min (WMM) policy, with the "weight" of each connection being its MCR requirement (we assume a nonzero MCR for each ABR connection). Furthermore, an explicit-rate (ER) based ABR switch algorithm is developed to achieve the WMM policy in the distributed network environment. Our ABR algorithm is proven to converge to the WMM policy through distributed and asynchronous iterations. The performance of our ABR algorithm is demonstrated by simulation results based on the benchmark network configurations suggested by the ATM Forum.

A generalized max-min network capacity assignment policy with a simple ABR implementation for an ATM LAN

Abstract: We introduce the generalized max-min (GMM) rate allocation policy, which is a direct generalization of the classical max-min policy with the support of both the minimum rate requirement and the peak rate constraint for each connection. A centralized algorithm is presented to compute network-wide bandwidth allocation to achieve this policy. Furthermore, a simple distributed algorithm with the aim of achieving the GMM policy is developed in the context of the ATM Forum ABR traffic management framework. The effectiveness of this distributed ABR algorithm is demonstrated by simulation results based on the benchmark network configurations suggested by the ATM Forum.

A simple ABR switch algorithm for the weighted max-min fairness policy

Abstract: An important concept in the ABR service model is the minimum cell rate (MCR) guarantee as well as the peak cell rate (PCR) constraint for each ABR virtual connection. Due to the MCR and PCR requirements, the well-known max-min fairness policy is not sufficient to determine the fair rate allocation in the ABR service model. We present the weighted max-min (WMM) fairness policy, which supports both the MCR and PCR requirements for each ABR virtual connection. A centralized algorithm is presented to compute network-wide bandwidth allocation to achieve this policy. Furthermore, a simple ABR algorithm based on the intelligent marking technique is developed with the aim of achieving the WMM fairness policy in the distributed ABR environment. The effectiveness of our ABR algorithm is demonstrated by simulation results based on the benchmark network configurations suggested by the ATM Forum.

Early selective packet discard for alternating resource access of TCP over ATM-UBR

Abstract: We investigate packet discarding schemes for TCP over ATM with UBR service. In doing so, we tested the effective throughput of two existing schemes, Partial Packet Discard (PPD) and Early Packet Discard (EPD), as compared to the Random Cell Discard (RCD) scheme which discards any incoming cells after buffer overflow. We observed that PPD alleviates the effect of packet fragmentation so that it gets effective throughput enhancement over RCD, and EPD provides further enhancement over PPD. After closer investigation, we found that there is a sustained congestion problem other than packet fragmentation that causes the effective throughput to be degraded. We noted that sustained congestion resulted in the synchronization of TCP window expansion and shrinkage. To provide a solution for this problem, we propose the Early Selective Packet Discard (ESPD) policy, a strategy which makes sessions take turns in accessing network capacity by discarding packets from selected sessions rather than randomly. Our results shows that ESPD achieves throughput and fairness enhancement over EPD with only a modest increase in implementation complexity.

Optimal space priority policies for shared memory ATM systems

Abstract: In this paper we study the problem of the optimal design of bu er management policies for a shared memory ATM switch or demultiplexer. A system with cells of two di erent space priorities is considered. Our objective is to determine the optimal policy that minimizes the total weighted cell loss. The problem of nding the optimal policy within the class of pushout and expelling policies is considered. Using sample path techniques the search space for the optimal policy is reduced to a subset of the entire policy set for each policy class. A numerical study based on value iteration technique is used to investigate the structure of the optimal policy. This technique is also used to calculate the loss probabilities for di erent classes of cells for a system with small bu er size.

Fair network bandwidth allocation with minimum rate guarantee and its ABR implementations

Abstract: A novel concept in available bit rate (ABR) service model as defined by the ATM Forum is the minimum cell rate (MCR) bandwidth guarantee for each connection. In this paper, we present a network bandwidth allocation policy to support each ABR connection's MCR requirement, as well as its peak cell rate (PCR) constraint. Furthermore, we develop two explicit-rate (ER) based ABR algorithms consistent with the ATM Forum ABR traffic management framework to achieve this rate allocation policy. The first ABR implementation is a simple heuristic algorithm which does not require per-VC accounting. It requires minimal implementation complexity and offers satisfactory performance in a LAN environment. The second ABR implementation employs per-VC accounting and is proven to converge to our rate allocation policy for any network topology and any set of link distances.

A rate allocation policy with MCR/PCR support and distributed ABR implementation using explicit rate feedback

Abstract: An important concept in the available bit rate (ABR) service model, as defined by the ATM Forum, is the minimum cell rate (MCR) guarantee as well as the peak cell rate (PCR) constraint for each ABR virtual connection (VC). Because of the MCR and PCR requirements, the well-known max-min fairness policy no longer suffices to determine rate allocation in the ABR service model. We introduce a network bandwidth assignment policy, MCRadd, which supports both the MCR and PCR requirements for each ABR virtual connection. A centralized algorithm is presented to compute network-wide bandwidth allocation to achieve this policy. Furthermore, an explicit-rate (ER) based ABR switch algorithm is developed to achieve the MCRadd policy in the distributed ABR environment and its convergence proof is also given. The performance of our ABR algorithm is demonstrated by simulation results based on the benchmark network configurations suggested by the ATM Forum.