One of our initial assumptions was that loose consistency would be acceptable on the World Wide-Web. This means that our protocols are optimized to reduce bandwidth and decrease latency, at the expense of occasionally returning stale data to the user. It is possible to remove the weak-consistency assumption by using server-initiated invalidation protocols at the cost of increased server load. This burden may become an obstacle to wide-area scalability.
As we saw in section 
, Worrell showed that it is
often impossible to accurately set TTL fields ahead of time in a
wide-area distributed system such as the World Wide
Web [37]. In an attempt to determine a context-independent
TTL he simulated the performance of a TTL-based cache. He also
simulated an Invalidation Protocol based consistency scheme, and
showed that the two schemes are essentially equivalent in terms of
their network bandwidth requirements when the default TTL is set to a
little over a week. At this point, however, one out of every five cache
requests are returned with stale data.
We hypothesized that a consistency mechanism originally designed for
use with caching FTP objects may also be useful on the World Wide
Web. The Alex cache consistency scheme [8] is based on the
assumption that the older an object is, the less likely it is to
change. This makes intuitive sense; an object that was just changed
may change frequently. An object that has not changed in several
months clearly changes infrequently. Specifically, the Alex scheme
uses an Update Threshold to decide when an object should be
considered stale. If time since last checked > total age 
update threshold then the file is considered invalid. The example
we provided in section was for an update threshold of
10%. In such a case, if a file is 1 month old, then alex will serve
the file for up to three days (
days = 3 days) before checking to see if it has become
invalid.