Improved Flow Aggregation in MPOA

• Main Author: 吳偉榮
• Other Authors: 林偉
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/36521025901733388249

碩士 === 國立中興大學 === 資訊科學研究所 === 88 === MPOA (Multi-Protocol Over ATM) incorporates the function of LANE(LAN Emulation) and NHRP(Next-Hop Resolution Protocol) to support intra and inter ELAN(Emulated LAN) communications over an ATM network. The objective of MPOA is to redirect the intersubnet forwarding path away from routers and onto the ATM switch fabric to speed up the communication. In MPOA, even at high cost with long delay for each shortcut setup, it can execute flow classification such that short-lived flow can hop-by-hop forward packet at layer 3 and long-lived flow can be switched at layer 2, so MPOA can offer significant performance gain over the traditional routed network in an inter ELAN communication environment. Packet flows in MPOA are classified according to their lengths. Long-running packet flows are identified and forwarded directly at layer 2 through shortcut paths. We combine several flows to avoid excessive operations for packet classification. Once a short-cut path is established, other flows with the same destination are permitted to move packets over the shortcut path without further classification. We find that when flows are aggregated, the average transmission rate of flows at layer 2 increases from 70% to 99%. Because the number of VC becomes fixed, flow aggregation effectively reduces the number of VC. This makes MPOA scalable. At present, AAL5 does not provide specific mechanisms for flow aggregation. A few aggregation methods have been proposed. However, these methods run into such problems as additional queuing delay, extra buffer and low cell utilization. Here, propose a method to improve the flow aggregation in MPOA. It uses a weighted round-robin scheme and performs flow aggregation. Our method improves cell utilization up to 90.5%. Cell of different flows can be interleaved for flow aggregation. This greatly reduces queuing delay and achieves higher throughput. The proposed method does not require additional buffers for flow classification.