home *** CD-ROM | disk | FTP | other *** search
- [ this is too big to put into 4.9.3 and still get it out on time;
- therefore i am putting it here in contrib/misc for now. i feel
- rather strongly that settransfer is the wrong approach to the
- problem it solves, but given that the code is clean and that there
- is an FYI out for it, i do plan to put it in as a default-"off"
- option as soon as possible. --vix ]
-
- Date: Thu, 1 Dec 94 20:04:14 EST
- From: brisco@hercules.rutgers.edu (Thomas P. Brisco)
- Message-Id: <9412020104.AA19390@hercules.rutgers.edu>
- To: paul@vix.com
- Subject: ``SETTRANSFER'' mods to 4.9.3.b10
- Cc: brisco@hercules.rutgers.edu
-
-
- Paul; included below (in shar format) are a set of diffs to
- implement the "transfer" keyword in BIND. These changes also include
- modifications to a few other areas -- you may want to look over
- README.TRANSFER and SETTRANSFER_CHANGES (the latter is aimed mostly
- at you). The changes file details as closely as I could the
- modifications I made to make the dynamics of subzones work as
- intended.
-
- You'll obviously want to look over the code rather closely, as
- there are changes in a few non-intuitive places. The #ifdef's also
- seek to make sure that ROUND_ROBIN and SETTRANSFER are mutually
- exclusive options - since ROUND_ROBIN tends to mess with the ordering
- of RRs, while SETTRANSFER needs them left in order. Similarly; all
- calls to sort_response() are ifdef'd out - as they also modify the
- ordering of the RRs. For "fast" subzones, I've also turned down
- maint_interval to 60 seconds (from 15 minutes) and min_cache_ttl to
- 60 seconds (from 5 minutes). I've actually ran subzones with 60
- second TTLs with no horrific side effects (other than increased CPU
- utiliziation :-).
-
- I've not included diffs for (a) the Makefile (ns_xfr.[co]
- needs to be added), and for (b) the conf.h file (though a
- "conf.h-additions" is included). In both of these, I thought
- you might have a preferable style, and since the changes are
- minimal, it's a bit briefer this way ...
-
- I'll be at the San Jose' IETF next week - are you going to be
- there? If so, I'll keep my eye open to say "hi". I'm hoping that
- the draft FYI will (finally) make it through, so I can lay this to
- rest. Though, if I do "get it to bed", I'll probably lend a hand
- with some of the other current activities for DNS (BTW: Thanks for
- putting me on the developer's list). Ah! Before I forget, this
- **does not** include the very latest change from Simon Leinen
- (re: BOGUSNS modifications).
-
- Tp.
-
- ========================================================================
- Any questions -> brisco@pilot.njin.net
-
- This is the PRODUCTION distribution of the SETTRANSFER code. The
- SETTRANSFER code permits the use of the keyword "transfer" to specify
- alternate programs to do the zone transfer if the current named is a
- secondary nameserver for a zone.
-
- The details on using alternative zone transfer mechanisms as a
- means for achieving load balancing is described in an internet rfc
- number 1794. This rfc is available from ds.internic.net via anonymous
- ftp, or via www (http://www.ineternic.net/
-
- Primarily the code changes adds enough syntax to do a database
- style lookup based on the zone name (nothing fancy, just an array),
- and to execute that program. Some necessary other changes were done
- in order to support ordered, quickly changing address records -
- primarily this had to do with lowering the value of static variables
- putting rather high minimum values in for the minimum zone refresh
- period, and minimum TTL associated with records. Additionally, in
- order to preserve the ordering of the records, all references to RR
- ordering are ifdef'd out. Also note that this, and Round Robin
- ordering are mutually exclusive. Code has been added to conf/options.h to
- ensure this.
-
- Specific changes are detailed in the file SETTRANSFER_CHANGES.
-
- Example configurations (and diffs to older versions of BIND) are
- available via anonymous ftp from pilot.njin.net:pub/TRANSFER.
-
- Currently (as of 8/28/95) there are two "fast" subdomains running at
- Rutgers University. One is eden.rutgers.edu running on the NS for
- eden.rutgers.edu ("eden" is a subzone of rutgers.edu). It uses the code
- from the LOADSAMP.shar to balance the load amongst seven SparcCenter 1000s.
- This is in active use in the Rutgers University domain.
-
- This version of the changes has been compiled by Robert Mende of
- Silicon Graphics Inc. Changes are primarly to keep the SETTRANSFER patches
- in sync with newer versios of named.
-
-