[bind10-dev] initial ideas on the "difference" design

[JINMEI Tatuya / 神明達哉]

At Fri, 14 Oct 2011 10:08:08 +0800,
"Likun Zhang" <zlkzhy at gmail.com> wrote:

> > In my original example of using a specialized table, the RRs might be
> > stored as binary data for performance reasons.  Right now I'm not sure
> > if we need such optimization, though.
> 
> My opinion is we'd better store binary data, not only for
> operformance, since the diff data is only cared by the
> slaves/masters, it will not be opened to end users. Why we need to
> en/decode them when doing tranfer-in/out?

(I'm quite neutral on the text vs binary per se, but) en/decoding
isn't avoidable with binary format either.  And, in fact, since we
already have routines for converting text<->RRset in the
database-based data source classes, the implementation cost would be
even cheaper for the text format.  As you said the diffs table would
basically be "captive" in practice, and in that sense we could more
easily justify the use of a binary format.  But, as the existence of
'named-journalprint' indicates, it would be convenient if the operator
can easily browse the diffs from the "journal", and for that purpose a
text format would be more handy (we can always provide a conversion
tool, but it would be costly both in terms of operation and
implementation).

Overall, I personally don't see the strong reason for adopting a
binary format if performance doesn't matter (while I see cases for a
text format).  And, actually, I guess performance might be an
important point if we consider high volume updates like thousands of
updates per second.

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.