[bind10-dev] recursor cache requirements - input required

Stephen Morris stephen at isc.org
Wed Dec 15 00:08:06 UTC 2010


On 14 Dec 2010, at 11:13, Likun Zhang wrote:

> Hi all,
> 
> I have updated the document of recursor cache according the comments from
> the bind10-dev list, the latest version is 8.
> 
> http://bind10.isc.org/wiki/RecursorCacheRequirements
> 
> There is one "Questions" section in the document, the only question is what
> the Key should be for the items in message/rrset cache
> 
> Shane's opinion is that only the class IN should be supported on the
> recursor cache, so the key for items in message/rrset cache will be
> "Query_name + Query_type". What shane said in his email is: Personally I
> really think we should just support class IN on the recursive side. CH is
> useful for getting ID.SERVER and the like, but nothing useful in the caching
> side, and the only other class is HS which MIT can send us a patch for. We
> can't get those 16 bits back from the protocol, but we can get them back in
> our implementation'' 
> 
> I will appreciate the comments on it from the reader.

I agree with Shane in that IN is by far the most common class, so storing an additional two bytes per query is just pure overhead.  However, a couple of points occur to me:

1) Suppose we get a query containing a class we have not seen before?
- Do we reject it?
- Do we handle it but not use a cache? (Will this be possible?)

2) Could we have a separate cache for each class?
- Should we create new caches "on the fly" as new classes are seen?
- Should be able to configure the classes we support?
(I prefer the second option - it avoids attacks whereby 64k caches are created to use up resources.)

Stephen






More information about the bind10-dev mailing list