On 10 Jun 2015, at 4:05, Davey Song wrote:
FYI, there is a thorough study done by Joe and Kim in June 2012 which is not included in the project overview document.
http://newgtlds.icann.org/en/about/historical-documentation/root-scaling-27j...
Thanks, Davey! I am particularly pleased that we managed to cite both Donna Summers and Angry Birds in Appendix E. That appendix provides describes a numerical model for predicting the compiled root zone size (using BIND9's named-compilezone as a normalisation tool) based on the number of TLDs and the prevalence of AAAA and DS records. It seems like fun to see how accurate it turned out to be. The numerical model is collapsed to three parameters, the number of TLDs (T), the number of AAAA RRs per TLD (alpha) and the number of DS RRs per TLD (delta). This gives S = 730585 bytes = 713 KB, or thereabouts. [walrus:~]% dig @xfr.cjr.dns.icann.org . axfr | awk '($2 ~ /^[0-9]+$/ && $3 == "IN" && $4 == "NS") { if (!tld[$1]) { tld[$1] = 1; T++; } } ($3 == "IN" && $4 == "AAAA" ) { aaaa++; } ($3 == "IN" && $4 == "DS") { ds++; } END { print T, aaaa/T, ds/T, T*(496.225 + 70.1988*aaaa/T + 125.387*ds/T); }' 964 1.7168 1.12552 730585 [walrus:~]% The actual size of the root zone according to the measure used in that report is 704 KB, so the model is off by 1.3%, which seems acceptable. Hooray! :-) [walrus:~]% named-compilezone -i none -f text -F raw=0 -o root.compiled . root zone ./IN: loaded serial 2015061000 (DNSSEC signed) dump zone to root.compiled...done OK [walrus:~]% ls -al root* -rw-r--r-- 1 jabley staff 953547 10 Jun 09:38 root -rw-r--r-- 1 jabley staff 720776 10 Jun 09:39 root.compiled [walrus:~]% Joe