Name: skT88420 Date: 11/16/99


Classic VM (build J2SDK.v.1.2.2:09/30/1999-19:00, native threads, jit_122)

The java.net.InetAddress class will keep adding CacheEntry objects to its
HashMap forever. This makes writing a crawler or any code that has to fetch
elements off the internet in a randomised way impossible with these classes.

As an example take the babelfish at "http://babelfish.altavista.com" which has
been rewritten as a servlet (to be released soon). This servlet needs to fetch
pages off the internet as requested by the users. Each request for a web page
translation adds one CacheEntry to the HashMap. But it is very unlikely that
after more that 15 minutes the babelfish will ever (in internet time) need to
retranslate that page. That element could therefore very well be cleared from
the cache.

The sun.net.inetaddr.ttl (undocumented) property added in JDK1.2 does not help
here as it only allows the programmer to set the time to live of a cache entry.
This tells InetAddress to look up that same address again if it comes across it
after the ttl period. But in the case that concerns me, the request may never be
made again. And my concern is not to make sure the inetAddress is accurate, only
that the memory consumed is reasonable.

InetAddress therefore needs to allow the programmer to set the type of Hash
table to use. The HashTable might have a limited size, or it might clean some of
its elements after a period of time, or whenever it decides to. It may want to
clean the oldest elements first or elements in a random order. Perhaps the
programmer should be directly able to access the hash table, if he is allowed to
do so by his security policy.

This problem would be even more clearly evident if someone were to write a
crawler in Java.
(Review ID: 97361)
======================================================================