On Thu, Apr 21, 2016 at 01:36:54PM -0700, Guyren Howe wrote:
> It's an enormous tragedy that all the development effort that has
> gone into NoSQL database has pretty much all gotten it wrong: by all
> means throw out SQL, but not the relational model with it. They're
> all just rehashing the debate over hierarchical storage from the
> 70s. Comp Sci courses should feature a history class.
This turns out to be true in many areas of language design, mutli-user
system security, virtually everything to do with networking, and
application deployment. I was at an IETF meeting some years ago where
someone talking about "Internet of Things" stuff was going on at
length about how nobody around the IETF really understood constrained
systems. Standing behind him at the mic was an assortment of
grey-bearded men who'd worked directly on the original IMPs (which
were 16-bit Honeywells that ran at like 5MHz and had IIRC 16Kwords of
memory).
It's also true that crappy interfaces that are good enough stick
around anyway. The UNIX Haters' Handbook is full of evidence of how
much less good UNIX was, but even Apple gave in. Also, many of the
historical compromises turn out, once you start to try to make
different ones, to be more obviously defensible. Most of the NoSQL
trend was not a hatred of SQL the language but a carelessness about
the relational syntax or a view that promises about consistency are
dumb. Then the first credit card number gets lost in an
eventually-consistent system, and people suddenly understand
viscerally why transactions semantics are so hard.
Best regards,
A
--
Andrew Sullivan
ajs@crankycanuck.ca