To get a sense for the likelihood that a request would take a long time to start receiving data, we collected 1000 random URLs from AltaVista [1] and timed their responses. This study differs somewhat from Viles and French [20], who studied the availability of random HTTP servers and the time to connect to them; here we are seeing how long it takes to collect the first data from a WWW page. Figure 5 shows the results of this experiment, based on the 722 URLs that returned data within the first minute. We found that about a third of pages responded within a second, assuming they responded at all, and half responded within about 1.6s. However, it takes 5s to cover of the pages and 10% took 10-30s or more for the first data to arrive. As more pages on the WWW are dynamically generated, we expect the fraction of pages with sluggish response to increase.
Figure 5: Distribution of response times to receive first data, based
on a sampling of 722 URLs. Roughly
of the sample received data within a second, and
received data within 5 seconds.