I appreciate the posting of this write-up, someone did a good job. I've often been curious how cache sharing was done at CF. I've tested this myself and found that it seems to take about 8 or so requests for the same record to get cached on all the nodes in a region before I stop seeing the requests show up at my DNS servers. I would be curious if they have a low cap on the TTL or have so many requests that requests are discarded quickly per FIFO.
Question for eastdakota, would you be willing to share any stats on cache hit/miss numbers and cache eviction rates due to the sheer requests per second? No worries if not I just think it would be interesting.