This project consists of a series of measurement-based studies of Web performance. Our ongoing activities include:

• evaluating the performance implications of different content serving architectures
• identifying and quantifying network factors that affect Web response time
• building tools to aid in Web performance measurement
• identifying the cause and location of network performance problems at hosting centers

Content Serving Alternatives

We have studied the impact of several alternative Web serving architectures on client-perceived response time

• single server location -- the traditional approach of placing content in a single data center that uses a single network provider (perhaps dual-homed with a secondary provider providing backup service).

• single server location with multiple ISPs -- in this approach the data center provides direct connectivity to multiple ISPs to improve performance and availability, usually with additional control over BGP routing to influence the ISP used to send packets to and from different clients.

• multiple server locations -- content is placed in few regional data centers to improve availability and performance for some clients

• widely distributed servers -- content is served from a content distribution network which has a large distributed set of small servers spread across many service providers

Network Factors and Web Response Time

End-to-end Web performance is influenced by a large number of factors including Web server platform, protocol features, and network characteristics. In contrast to earlier work that primarily studied the impact of Web servers or the HTTP protocol on Web performance, this project examines network-related factors. In particular, we focus on the impact of network path characteristics and network connectivity of both clients and servers on Web response time. Our findings are based on two sets of measurements from diversely connected measurement agents to two identical servers placed in commercial hosting locations.