Check out the new USENIX Web site. next up previous
Next: Summary Up: Performance Comparison Previous: Reduced Latency

Network Costs


To incorporate network cost considerations,
GD-Size(hops) sets the cost of each document to the hop value associated with the Web server of the document, and GD-Size(weightedhops) sets the cost to be . Figure 7(b) and 7(c) show the hop reduction and weighted-hop reduction for LRU, GD-Size(1), GD-Size(hops), and

The results show that algorithms that consider network costs do perform better than algorithms that are oblivious to them. The results here are different from the latency results because the network cost associated with a document does not change during our simulation. The results also show that the specifically designed algorithms achieve their effect. For hop reduction, GD-Size(hops) performs the best, and for weighted-hop reduction, GD-Size(weightedhops) performs the best. This shows that GreedyDual-Size not only can combine cost concerns nicely with size and locality, but is also very flexible and can accommodate a variety of performance goals.

Thus, we recommend GD-Size(hops) as the replacement algorithm for the regulatory role of proxy caches. If the network cost is proportional to the number of bytes or packets, then GD-Size(weightedhops) is the appropriate algorithm.

Pei Cao
Thu Oct 23 18:04:42 CDT 1997