Summary

Asia-Pacific Network Operations and Management Symposium

2014

Session Number:TS8

Session:

Number:TS8-2

Path Capacity Estimation by Passive Measurement for the Constant Monitoring of Every Network Path

Naoyoshi Ohkawa ,  Yuji Nomura,  

pp.-

Publication Date:2014/09/17

Online ISSN:2188-5079

DOI:10.34385/proc.21.TS8-2

PDF download (375.4KB)

Summary:
Degradation of networking quality is a serious problem for service providers (e.g., on-demand video subscription service) since it causes customer defection in their services and results in a decline of their sales. Insufficient path capacity is one of the typical causes which degrade networking quality. Especially, the recent technical trend to shift virtualized network, in which the network is dynamically and autonomically reconfigured, increases the risk of unexpected insufficiency of the path capacity in the network. To detect such an unexpected insufficiency of path capacity, a constant monitoring of the whole network is highly expected. Path capacity estimation with passive measurement method is an approach which suits to such a purpose since it adds no additional load to the target network. However, existing passive measurement methods tend to either 1) have insufficient accuracy on their estimation or 2) require heavy computational cost for their histogram analyses. In this paper, we introduce a novel path capacity estimation method by passive measurement for a constant monitoring of a network 24-hour, 365-day. Our method can estimate the capacity of each path with sufficient accuracy by eliminating the two factors of degrading the estimation accuracy, which are the influence of TCP window flow control and the influence of cross traffic on the path. We evaluated our method using packets captured from our in-company backbone network. Our method accurately estimated the capacity of the narrowest link in every 1524 connections, in which 80% of the connections were within approximately 15% of their actual values (in over 1MB traffic case), as shown in Fig.8.