We have also incorporated new features to WPB to generate bursty traffic, with peak loads that exceed the capacity of the proxy, based on [1]. We have changed the client process code to have a pair of process - sender and handler - generating HTTP traffic to the proxy using non-blocking sockets. The sender process is responsible for establishing a connection with the proxy. After the connection is established, it passes the socket descriptor to the handler and sends a new connection request. The handler is responsible for keeping track of all pending connections and receiving the requested files from the proxy. If the connection could not be established in a reasonable amount of time (500 ms in our implementation), the sender issues a new connection request immediately. Each pair sender-handler manages a pre-defined number of sockets descriptors so that the number of sockets (simulated clients) that are trying to establish new connections is kept constant.
We have done some experiments with this new version of WPB (version 1.1) using different proxies. The main results show that there are a lot of connection time-outs that are reported as errors. The timeout period used in our implementation (500 ms) is much larger than the maximal round-trip time between client and proxy in our testbed (described in section 3). Therefore, the proxies were actually overloaded and could not handle many of the requests. For some proxies, we observed a very big number of errors so that the effective proxy throughput was very low. As a consequence, latency dropped very significantly. We are still working on this new version of WPB and trying to interpret the results. Therefore, all the results presented in this paper were collected using version 1.0 of WPB.