Multihoming Case Study

B. -Multihoming Improvements
To understand performance benefits of -multihoming, we adopt the following simple methodology: For each download, we compare the client-perceived turnaround time achieved by using the best ISP among all those available in the city, with that from using the best ISP in a candidate multihoming option. We average this ratio over transfers to all clients, and report the minimum normalized performance metric (the minimum is taken over all candidate options).We compare only those transactions for which there was a successful transfer over all ISPs at roughly the same time. In Fig. 2, we plot the above RTT and throughput metrics due to -multihoming as a function of the number of ISPs. Two key facts are apparent from Fig.
Practical Route Control

So far, we studied the potential improvements from multihoming by analyzing an ideal form of multihoming that was driven by certain key assumptions: First, the end-network had perfect information of the performance of ISP links for each destination. Second, the end-network did no incur any overhead in moving traffic across ISPs over time. Third, the end network was able to control the ISP link taken by traffic entering the network. To realize these potential performance benefits in practice, we must address the following issues:
(1.a) How should end-networks monitor the per-formance of ISP links? Is active probing better than passive observation?
(1.b) A related question is which destinations to monitor. Should the end network probe all poss-ible destinations via each ISP link? Does this give rise to scalability issues?
(2) How should the end network estimate the future performance of an ISP to a destination? This is key to determining which ISP the end-network must use for the destination.
This choice depends on the time-varying performance of each ISP link to each destination being accessed. However, net-work performance could fluctuate, very substan-tially on some occasions [10]. A multihomed enterprise, therefore, needs effective mechanisms to monitor the performance for most, if not all, destinations over each of its ISP links. There are two further issues in monitoring performance over ISP links: what to monitor and how. In the enterprise case, one would ideally like to moni-tor the performance from every possible content provider over each ISP link. However, this may be infeasible in the case of a large enterprise which accesses content from many different sources. A simpler, more scalable solution to this problem is to monitor only the most impor-tant or popular destinations. But, how can we track the most popular destinations in a scalable and efficient manner, while also accommodating temporary shifts in popularity (e.g., due to events such as flash crowds)? In Section IV, we outline several approaches to track the popularity of destinations in a timely and scalable manner.
For the second question, two common ap roach-es are active and passive monitoring. In active monitoring, the multihomedenterprise performs out-of-band measurements to or from specific