Re: [DotGNU]Potentially useful kit

[Bill Lance]

--- Chris Smith <address@hidden> wrote:
> On Monday 07 January 2002 18:20, you wrote:
> >
>
http://subscribe.dotgnu.org/pipermail/developers/2002-January/001636.html
> >
> > It sure sounds to me like a lot of what you've
> done
> > with Goldwater fills in a lot of detail in the
> Vritual
> > DataServer idea.
> >
> > What do you think?
> 
> On first reading, the distributed nature looks
> pretty similar.  You could 
> certainly model the behavior you propose with
> goldwater. 
> 

The Virtual DataServer is just a very rough concept at
this point.  It has not been pounded on or discussed
in any detail yet. Barry Fitzgerald has expressed
interest is a somewhat overlapping project as well. 
His focus is more on a traditional ASP.  The VDS idea
is more focused on eliminating a central server
entierly, while preserving some of their advantages. 
Those advantages being:

  *  persisant and robust presence on the net

  *  a common and managable pool of data
 



> However where Goldwater has a service directory (or
> pin-board) where 
> Goldwater service-processes post their service
> lists, this cannot be made to 
> model the Dataset Repository you in your spec.
> 
> However, an idea occurs to me! Yay!
> Goldwater service-processes are just regular
> executable or Perl scripts. 

Is it written in PERL?
 
> Now it is entirely possible to write the business
> logic implementing the data 
> repository you describe as a collection of Goldwater
> services.  That's the 
> sort of thing the middleware was designed for.  It
> also becomes VERY modular, 
> making enhancements and maintenance very
> straightforward.
> 
> The goldwater clustering would then take care of
> inter-repository data 
> sharing and workstation subscriptions etc.
> 
> At least I think so....

This is interesting.  I was thinking of something like
FreeNet, but there are several problems with that. 
One is the high overhead for what may be vast overkill
for anonymity and perhaps some for security.  Also,
data management in Freenet is tricky,  some would say
impossible.

> 
> Oh, we'd need to abstract the call to the
> appropriate Goldwater service from 
> the dotGNU call, which I'm assuming will be a SOAP
> request, Yes ???
> 

and XML-RPC, and perhaps J2EE as well,


> I'm unsure as to how discrete VRS's know about each
> other - will this be 
> fixed configurations, setup when a VRS is
> commissioned?  Or are you 
> describing a 'star' topology where a central
> 'master' VRS (or even a cluster 
> as they can be centrally managed) is known about by
> all remote VRS's, and the 
> remote VRS's announce their existence when they
> subscribe to the 'master' 
> VRS's???
> 
> Also, I'm also a little unsure of is how the LDS
> fits into all of this, 
> <thinking on his feet...> unless the LDS is in fact
> another Goldwater service 
> to which a dataset is forwarded, the result from
> which is returned back to 
> the client.  Yeah, that might work.
>

The Local DataServer is the only actual piece of
software running, on each individual cluster member
machine.  The Vitrual DataServer emerges from the
collective behavior of the cluster of the member
LDS's. It is not a seperate piece of software, itself.
 That's why I call it a 'Virtual' server.


Internal to a cluster, each member knows about all
other members through the subscription process.  A
subscription to a cluster is persisant, so each member
can have a list of all members IP's.

Access to cluster from the outside is a different
matter.  First, access to the cluster as a whole is
through any one member.  Conversly, the cluster can
make it'self available to to outside net by providing
any one or a group of member IP's.  There a a number
of ways to do this, depending on how public the
cluster wishes to be.

 
> Hmmmm.  What do you think?
> 
> BTW - I'm wondering now what platforms the VRS is
> intended to run.  There are 
> some major parts of Goldwater that are primarily
> Unix only - shared memory 
> and ipc etc.  Phoenix should port to M$ easily
> enough - it was originally 
> designed to, but never done really.
> 


As crossplarform as possible.  This may depend on just
how crossplatform pnet ends up being.  


__________________________________________________
Do You Yahoo!?
Send FREE video emails in Yahoo! Mail!
http://promo.yahoo.com/videomail/