Check out the new USENIX Web site.
... URL1
In Section 3.3 we explain why it may be useful to use one of the physical URLs as a logical URL.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... 4.02
Both browsers executed on a PC with 300 MHz Pentium II processor and 64 MB of main memory running Windows NT Workstation 4.0.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... system3
If a shared file system is not available, each client uses its local version of latency table.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... latency4
In all experiments we measured also HTTP response time and found its behavior fairly close to that of HTTP request latency.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... period5
We did not use a higher polling rate as it could be interpreted as a denial-of-service attack.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... machine6
Sun SparcStation with 128 MB of RAM running Solaris 2.5 and Java Web Server 1.1.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... interpreter7
Some of the clients ran on platforms that did not support JDK1.1 at the time of experiment.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... timeouts8
In most cases the calculated timeout was larger than the timeout of the underlying java.net.URLConnection implementation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... client9
The DNS client may be in a completely different location than the Web client if a recursive DNS query is used
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.