We have measured the performance of two commercial content distribution networks (CDNs), one operated by Akamai and one operated by Digital Island. Although there are differences in the implementation of these services, both CDNs redirect requests by using DNS. In this paper, we describe our simple measurement technique for a DNS-based CDN, our data for the two commercial services, and our interpretation of that data. Our main conclusion is that CDNs provide a valuable service, but that neither Akamai nor Digital Island can consistently pick the best server of those available. Contrary to some widely disseminated marketing messages, we argue that CDNs succeed not so much by choosing an ''optimal'' server as by avoiding notably bad servers.