ELINSKY@YKTVMZ.BITNET ("Jay Elinsky") (01/16/90)
I'm posting this for someone who has read-only access to this list. He
asks that you post comments to the list.
I know of two different methods for implementing a resolver, which are
described below. I'd like to hear comments on which is preferred
because of usability, and any problems with either of the schemes.
Both require a RESOLV.CONF file that contain the standard DOMAIN
statement and NAMESERVER statements.
Example: DOMAIN WATSON.IBM.COM
NAMESERVER X.X.X.X
Method I
--------
If the host name does not contain a period, append the domain name
before sending the name to the resolver.
Example: PING YORKTOWN sends YORKTOWN.WATSON.IBM.COM to the resolver.
PING YORKTOWN.WATSON sends YORKTOWN.WATSON to the resolver.
Advantages: Simple to implement, simple to understand, only one
query is ever sent to the name server.
Method II
---------
Always append the domain name, and gradually remove the first element
in the DOMAIN statement trying all possibilities.
Example: PING YORKTOWN.WATSON.IBM.COM
would send to the resolver: YORKTOWN.WATSON.IBM.COM.WATSON.IBM.COM
then: YORKTOWN.WATSON.IBM.COM.IBM.COM
then: YORKTOWN.WATSON.IBM.COM.COM
then: YORKTOWN.WATSON.IBM.COM (success)
PING YORKTOWN.WATSON
would send to the resolver: YORKTOWN.WATSON.WATSON.IBM.COM
then: YORKTOWN.WATSON.IBM.COM (success)
Advantages: A user could specify part of his domain (like
YORKTOWN.WATSON), provided that part was in their domain
tree.
Disadvantages: Complicated, the Hosts Requirements RFC says that if
this method is implemented that a method must be provided
to turn it off, and that all negative queries must be
cached to prevent further negative traffic to the DNS.
Has anybody successfully implemented Method II? And if so, what kind
of user feedback have you gotten?
Daniel M. Barton
TCP/IP Development
IBM Corporation
Research Triangle Park, NC08071TCP@MSU.BITNET (Doug Nelson) (01/17/90)
> I know of two different methods for implementing a resolver, which are > described below. I'd like to hear comments on which is preferred > because of usability, and any problems with either of the schemes. > > Method I > -------- > If the host name does not contain a period, append the domain name > before sending the name to the resolver. > > Method II > --------- > Always append the domain name, and gradually remove the first element > in the DOMAIN statement trying all possibilities. > > Has anybody successfully implemented Method II? And if so, what kind > of user feedback have you gotten? I have edited out the examples from Dan's comments above. We don't use method II, but my biggest concern with it is the potential for false hits, or the potential for hiding a domain name. For example, say I have a "com" (e.g. Communications) department. And say they name their Big Blue system "ibm" (for lack of creativity or whatever). Its full host name is thus "ibm.com.msu.edu". Now what happens when a reference is made to "ibm.com"? Should this be my local system, or the top-level entry for your company? I think it has to be the latter. In other words, any name with a dot in it should *always* be tried first as a fully qualified name prior to appending any portion of the local domain or domain(s) from the DOMAIN list, or to removing part of the requested name. If this were done, I could accept a method II implementation. I'd suggest it be controlled by a keyword in the resolv.conf file (ALWAYSAPPEND). Another method that I have seen is a variation on method II, where dotted names are always tried as is, but single-part names are tried as in method II (in domain a.b.c, first try host.a.b.c, then host.b.c, then host.c). I have aided that process here by the fairly liberal use of CNAMEs - each host system, and a number of workstations, named as "host.dept.msu.edu", each have a CNAME entry under "host.msu.edu". Another useful aid is to place more than one DOMAIN statement in your resolv.conf file, giving a list of your favorite domains. Doug Nelson nelson@msu.bitnet Manager of Network Software nelson@msu.edu Michigan State University