Provisioning for bursty Internet traffic: Implications for industry and Internet structure
David Draper Clark, William H Lehr, Ian Liu · 2000
Internet traffic, as epitomized by Web browsing behavior, is very bursty, or equivalently, the ratio of the peak to average data rate is quite high. To handle the offered traffic, the network must be sized to handle the peak load. Because the peaks of individual users are typically uncorrelated, the network peak load grows much more slowly than the sum of the peak loads of the individual subscribers whose traffic is carried by the network. This implies there are provisioning scale economies associated with aggregating traffic. That is, service providers that are able to aggregate the traffic of a larger number of users are likely to have lower capital and operating costs and there may be a minimum efficient scale of operation for Internet Service Providers (ISPs). In the current environment where users are accustomed to a relatively poor grade of service (i.e., long packet delays are tolerated) and when most users access the Internet via dial-up modems that limit the possible peak to average load ratio, the impact of these scale economies on industry structure are likely to be small. However, with increased quality of service (QoS) expectations (e.g., addition of delay-intolerant real-time services) and the spread of broadband services offering the potential for much higher peak to average load ratios, the provisioning problem may grow in importance. Although models of aggregate traffic flows have been developed [Kelly, 1998] and others have speculated about the nature of interconnection agreements across the Internet hierarchy [Bailey and McKnight, 1997; Lehr, 1998], we are unaware of any work that attempts to argue from traffic characteristics to industry structure.