Re: One for Keith

Ron Resnick (
Sun, 21 Jun 1998 13:13:27 -0400

Poor Keith. Having his name dragged through all this. He's
got nothing to do with any of it :)

Joe Barrera wrote:

> Oh, come on. There are RTF viewers and editors for every platform I know
> about. And again, the conversion to HTML is straightforward (although
> lossy).

Hmm. No you missed my (not very profound) point. RTF is an MS
standard. Sure, it's been ported everywhere. I was just having a little
fun with "god fearing RTF", turning the phrase around a bit.

> You know, I think one of the things I'm going to end up trying to do in my
> career is reconciling the Ken Birman virtual synchrony view of the world
> with the Jim Gray transactional view of the world.

May I suggest starting with
Transaction Model vs Virtual Synchrony model: bridging the gap
Rachid Guerraoui, Andre Schiper,
Ecole Polytechnique Federale de Lausanne
(Hmm. I just checked, and couldn't find it online - just
a Bibtex reference. If you want, let me know and I'll email
you the .ps file).

That LSE bunch is way out in front on this stuff, really
leapfrogged Cornell I think. I have my own views on VS/TP
which are more hand-waving, a lot less mathematically based
than the above paper's formal treatment (reduction to a common
DTM model). Basically, much of my argument has to do with
the inherent pessimissm/optimism in the approaches. The rollback
vs. make-forward-progress thing. TPs will take a vote amongst
a dozen members, decide someone's not happy, and roll back.
VS tries to keep going while this is at all possible, and still
preserves safety constraints. It's a difference in outlook, I think.

> (Why me? just because I
> work for Jim in a small lab, and Jim's interested in the question, and Jim
> and Ken are friends, and because the question interests me. On the other
> hand, Werner probably has it all figured out already -- I'm just constantly
> amazed how sharp he is.)

Aye, he is. As are the other sharp cookies in this business -
the Dougs (Lea & Schmidt), Silvano, the LSE crew, van Rennesse, etc..
I don't claim to inhabit that stratosphere- I just want to apply their
stuff where the masses can get at it.

[Snipping most of your treatment, not because it's not interesting,
but because the only point I want to address right now is:]
>So replication
> is still there, it just gets pushed down much lower.

This is a very common approach. I think it's one Birman&crew
take themselves way too much. Groups are hard, they're system
programming level constructs, they're not meant for application
level APIs. So lets bury them and slap "simpler" interfaces
over top. I don't buy this. I think the group (session, channel,
community, shippable place) whatever is perhaps the _fundamental_
user-vsible abstraction, right after the granular objects/documents/bits
themselves. One of the most influential papers I've ever read
(and love citing every so often to drive home how important the
ideas are) is:

Objects in Groups, Doug Lea

Groups are *not* just replication and FT.

I would really suggest switching this to dist-obj if you want
to discuss this further.