dns and software, was Re: Reliable Cloud host ?

David Conrad drc at virtualized.org
Thu Mar 1 13:11:51 CST 2012


On Mar 1, 2012, at 10:00 AM, Michael Thomas wrote:
> My rule of thumb is for this sort of thing "does it *require* kernel level access?"
> In this case, the answer is manifestly "no".

This is tilting at windmills since it's wildly unlikely anything will change, but...

The idea is to add a level of indirection that does not currently exist, similar to the mapping of filename/file handle/inode in the filesystem.  This layer of indirection allows the kernel to remap things as it sees fit without impacting the application.  If such functionality existed, the kernel could manage the mapping between name and address to do things like honoring DNS TTL, transparently handling renumbering events, deal with protocol transitions even during a connection, etc.  As things are now, it's like having to rewrite non-tivial sections of code for _all_ disk-aware applications because we've gone from a 32-bit file system to a 64-bit file system, even though the vast majority of those applications couldn't care less.

> SRV records?

Do not have addresses in their RDATA, they have names.


More information about the NANOG mailing list