Re: [Gnu-arch-users] Re: RFC: arch protocol, smart server, and tla implementation prototypes

[Tom Lord]

    > From: Chris Gray <address@hidden>

    > > Yeah, but you made up a stupid server implementation that would have
    > > n^2 behavior and you're trying to use that to argue against the
    > > protocols and architecture that would fail to prevent you from
    > > implementing such a stupid server.  Can I have some more of that,
    > > but pickled and on a cracker, please?

    > Oh, because it's a red herring.  It's funny because it's a cliche.
    > Well as soon as you can fill my pudding bowl with this server that
    > knows when to cache things, I'll accept that as proof.  

    > I guess I blundered when I brought up the DoS attack though.  That
    > sort of obscured my original message.  My message was that skiplists
    > are a data structure that you should consider for reasons of
    > efficiency, simplicity, and size.  I feel they would be preferable to
    > what we have now and easier to implement than a truly smart server.


I look at it this way:

Our back and forth sniping about the quality of your criticisms and
the quality of my criticisms of your criticisms -- that's just
carrier.  Mildly entertaining when we nail it and just silly when we
don't.  It's also "provocative carrier" in the sense that it inspired
some new contributions to the thread that contained some good signal.

Meanwhile, somewhere off to the side, is good old Colin -- working on
his protocol and his prototype server.   One of expressed goals for
the server is to make it very "pluggable" and (my interpolation) just
generally easy to hack on.

How can anyone ever hope to do that?  Design a piece of software in a
new domain but keeping in mind all the ways that people might want to
customize/improve it down the line?   It's impossible to do
perfectly.   He (and anyone else working in a similar area) can at
best be exposed to lots of possibilities, roll them around in the
mind, then do the simplest thing that still makes sense.

Our little flame war (if it even warrants that name) imposed some
signal on the flamish carrier.  An informal sketch of the design space
and the options and pointers towards how to think about the
trade-offs.

Nowhere in any of that was the skip-lists idea refuted.  On the
contrary, if you read closely: we agree on it as a strategy for smart
servers and as something that can be done with explicit
cache-delta'ing on dumb-fs servers.  All the flamage was just about
the noise that surrounded the important and true points in the thread
-- to which I think just about everyone who participated contributed
some.

So: thanks for raising the point.  Now, let's look forward to the
pudding.

-t