[vworld-tech] Ultimate MMO Platform

[J C Lawrence]

On Mon, 12 Apr 2004 02:06:42 +0100 
Dave  <fear at planetquake.com> wrote:
> J C Lawrence wrote:

>> Wouldn't the notions of quorum mechanics, such as the degenerate
>> cases where both ends of a transaction cross-check/sanity_check the
>> results of a computation or current game state directly contradict
>> this?  They are 'consensual systems' which yet offer (potentially)
>> strong definition and correctness controls[1].

> Doesnt quorum mechanics rely on the fact that some nodes are more
> reliable than others? 

Not necessarily.  The usual case is for N other nodes to
cross-check/duplicate/sanity_check work done with disagreements among
the nodes being considered signal of errors with appropriate rollbacks
or exclusions being done.  The size of N can vary, as well as the
randomity of the assignment of which other node(s) will be a part of N.

> I was thinking of a system similar to a web of trust like PGP.

Right, and I was specifically contrasting to those systems which assume
untrusted and untrustworthy nodes, and yet maintain logical consistency.
PGP assumes that private keys are inviolate and trustworthy.  The whole
system breaks down if that assumption can't be maintained.

  Compare the current SSL key situation were the private keys of one or
  more of Verisign, Baltimore Technologies and Thawte to be publicly
  posted.  The whole system instantly collapses: SPOF.

> Another point about using peer to peer is how you prevent users from
> extracting more information than they should about the area of the
> world that they are running?

There have been some interesting discussions on that line on MUD-Dev in
the last months, in particular of the idea of distributing the full
world set/stream in encrypted form, but only distributing the keys
just-in-time to a specific node that it needs to access the relevant
data at that time.

> Someone you once considered reliable may not be anymore, intentionally
> or not. This of course does cost in redundant operations but I suppose
> like everything it is a compromise of security vs performance.

FWIW this area (trust relations in knowledge graphs with untrusted
nodes) is an area of considerable worth in the crypto and protocols
fields with some really good work done in the last few years.  CiteSeer
and Google should take you to the relevant papers.

> IIRC machines like the SGI Origin use hypercubes (is that right?) that
> would be an excellent way of organising p2p networks. 

SGI Origins are just single machines.  Like most SGI boxen they can be
organised into NUMA clusters via HIPPI or GSM interconnects.  Hypercubes
are one possible arrangement of nodes in such a setup.  I know that
other patterns are used (eg with smaller and larger node counts), but I
don't know the details.

-- 
J C Lawrence
---------(*)                Satan, oscillate my metallic sonatas.
claw at kanga.nu               He lived as a devil, eh?
http://www.kanga.nu/~claw/  Evil is a name of a foeman, as I live.