This paper looks at DNS traces from two locations, MIT and Korea, to figure out how well DNS does at finding names, what distribution user requests have, and what kinds of errors occur. The surprising result is that about 23% or requests get no response and 13% get negative responses. Interestingly, many of the latter were due to reverse mappings not existing. It would seem that not many applications use reverse DNS, but it might well be hidden in system calls, libraries, etc that applications do use. This is one issue with running a system used by libraries and applications like DNS - you might never see what the applications are actually doing.
The paper also looks at whether more aggressive caching would improve DNS performance and decides that it won't due to the Zipifian distribution of results. However, this is done for caches of 10-20 clients - I'd think that ISP level caches can help significantly more.
One question I have is why are we worried about DNS performance? Was it more of a problem when this paper was written? DNS seems to work pretty well for me for almost all sites I want to get to. It's good to see a thorough analysis of it but I don't fully understand what the motivaton for this was.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment