CoDoNS

This paper studies an alternative to the existing DNS service, called
CoDoNS, which builds atop a DHT overlay. By combining Pastry with
Beehive (a pro-active caching mechanism for reducing search times) they
are able to achieve good load balancing and failure resistance along
with comparable (if not better) response times.

Under the DHT assumption that all nodes are homogeneous in capacity,
CoDoNS performs well, able to distribute load fairly evenly, recover
from failures, handle updates, and service popular questions quickly
(often faster than legacy DNS).

Of course, it is not the case that DNS servers are homogeneous in
capabilities. The current system provides a natural incentive for
distributing server capacity -- popular servers naturally need to be
bigger and more robust. My own personal DNS server, for example, is a
Pentium 120. I'm not sure why it's all that desirable to make my P120
serve DNS queries for other people.

The paper also argues that by flattening the architecture of DNS
(though still maintaining the hierarchical naming structure),
namespace owners have greater control over their names. They cite
malicious alteration as a motivation for CoDoNS. However, CoDoNS
relies in a central certification authority. This method has two
weaknesses: first, it places power in the hands of CAs. Perhaps not
to discriminate per-name, but certainly to discriminate financially
(e.g. expensive Verisign certs). Second, there is no reason that
legacy DNS cannot be retro-fitted with signed records so that entries
cannot be maliciously modified.

On page 4, column 1, paragraph 1, they state that it is surprising to
find 0.82% (less than one percent) of namespaces served by only one
nameserver. I am not sure why this is so surprising. If a Mission in
a remote part of Congo can only afford to buy and power one DNS
server, I don't understand why this is so shockingly surprising. And
if that DNS server in the Congo goes down, what's the big deal?
Chances are the server the DNS points to is down too. I think they
have bigger problems to worry about than DNS.

The paper also argues against legacy DNS by citing the existence of
many unpatched (but well known) vulnerabilities in existing DNS
servers. But I find this argument weak because CoDoNS does not in any
way guarantee participating name servers will be any less vulnerable.
In fact, assuming that more nameserver systems join, we expect more of
them to contain unpatched vulnerabilities. A sufficiently large
number of compromised CoDoNS nodes can maliciously coordinate and
disrupt service.

Like Chord DNS, CoDoNS also suffers from lack of control over a
namespace once it is injected into the system. UofT maintains and is
the authority of its own names. Under CoDoNS, UofT maintains its
namespace via crypto-signing, but the home-node for UofT's names are
likely at different nodes. At the same time, UofT may find itself the
designated home for animalfetish.com. This isn't so much of a
technological problem as it is a political and administrative problem.
Received on Wed Nov 09 2005 - 16:35:28 EST