Improved local DNS configuration

For the client and local DNS associations that are not in the same network cluster, we ask whether there exist any local DNS servers in those clusters. From our log, we collected a set of local DNS servers. Thus, assuming the clients have access to those local DNS servers in their network clusters, it is interesting to examine the degree of improvement if all LDNS servers were used optimally. This assumption is not unreasonable, since most IP addresses in the same network cluster are under the same administrative control. From Table 4, we can calculate the number of client ASes and network clusters where there are no local DNS servers as observed in our log. There are $9,570-8,590=980$ such AS clusters, and $104,950-53,321=51,629$ such network clusters. Table 7 compares the improved percentages of client-LDNS associations and HTTP requests in the same cluster with the original results. If the clients in our data currently configured to use a LDNS in a different cluster are allowed to use an LDNS in the same cluster, then at least 92% of the HTTP requests come from clients using the LDNS in the same AS cluster. That number is 70% for network clusters.

Table 7: Improvement of the percentage of the client-LDNS associations sharing the same cluster using optimal LDNS assignment

Metrics	Client IPs		HTTP requests
	Original	Improved	Original	Improved
AS cluster	64%	88%	69%	92%
Network cluster	16%	66%	24%	70%