Search Engine Performance Comparisons - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 160

A few facts emerged from the tables. Both Yahoo! and AlltheWeb have similar access timing from
the east coast and from the west coast. For AlltheWeb it was an average of 636 milli-seconds from the
west compared to 606 milli-seconds from the east (less than 5 percent difference). For Yahoo! it was
573 milli-seconds from the west compared to 627 from the east (about 10 percent difference). For MSE,
it was 355 milli-seconds from the west compared to 480 milli-seconds from the east (about 35 percent
difference). This is an indication that MSE has not replicated its service evenly throughout the country,
as some other search engines do (e.g. Google). The second observation is that there is a clear difference
in time needed to serve the first page and the rest of the pages. This can be explained by the reason that
when the search engine receives the query, it has to retrieve relevant pages from vast amount data. Once
retrieved, the subsequent pages can be cached, which speeds up the service time. The third fact is that
MSE has overall the fastest average service time.
Since the timing statistics were taken from two different computers, one from Bucknell University
on the east coast and the other one from California State University at Los Angeles on the west coast.
The hardware and software systems on both computers also have an effect on these timing. In order
to minimize this side effect, a baseline timing is collected. Each of the two computers was instructed
to access a local Web server installed on the host computer. The local Web server generates a random
page for each given query, simulating the behavior of a real search engine. The same number of Web
pages was requested as if it were accessing a chosen search engine on the Internet. The baseline timing
statistics are presented here in Table 8.
As can be seen from Table 8 that the two desk-top systems used in the measurement have quite
different timing statistics, indicating two quite different computer systems used in the experiments. It
is not practical to have the exact, completely un-biased comparison from the two sites because of the
differences in hardware and software used on the desk-top computers and in the network environment
on the two campuses. We attempted to minimize the discrepancy by subtracting the local access time
from the overall statistics and compare the results.

measured timing = overall timing - local access timing

Table 8. Baseline wall-clock time (in milli-seconds)

Search Engine Local host (east) Local host (west)

Avg. Page 1 108 45.4

Std. Dev. Page 1 7.34 9.82

Avg. Pages 2-5 13.7 5.41

Std. Dev. Pages 2-5 0.36 1.39

Avg. Overall 60.7 25.3

Table 9. Adjusted wall-clock timing for a page seen from Bucknell University (in milli-seconds)

Search Engine AlltheWeb MSE Yahoo

Avg. Page 1 587 499 612

Avg. Pages 2-5 505 341 520

Avg. Overall

545 419 566