[ANet-devel] Re: Basic requirements

[Benoit N. <be...@ma...>]

>I think we're getting mixed up here between the message protocol and the needs
>of applications built upon it.  If you have a generic anonymous message
>library
>that allows you to send a message between two nodes of the distributed network
>then you'll be able to use it to send queries or data transfers.

Not data transfers, as somehow two nodes must become proxies that
communicate directly with each other (IP to IP), and not through the
network. But the data that is transfered from the node to the proxy is
transfered like queries.


>I'm thinking about the mechanics of moving a piece of data around the network
>no what the nodes do with the data once they have it.

Ooohhhhh... You mean the "packets" of data that will be moving around the
network?


>The query is a broadcast message - please can I have a file called X.

You got it.


>Static data is a piece of a file - please transmit this data from A to B.

Uh... No. "Static" means that some data stays for a long time in the
network. Maybe I wasn't clear enough about that. So, unless you want a file
to "stay" in the network (with cache systems), you never use static data,
especially when you just want to transfer some data from A to B. "Static
data" means that the data will be distributed to the whole network (like
queries), but is bigger than queries and will stay in the network a longer
amount of time. "Two way data transfer" means that two nodes want to
exchange between each other and that the data should neither be cached nor
duplicated to the whole network. Those two nodes can be any node on the
network, not just two nodes that are adjacent that are connected to each
other, IP to IP, as part of the network.

What I'm trying to do here is to define the way ALL "packets" of data
should behave on the network, so that any service you can think of can work
on top of that. Think about what happens with ports in TCP. You don't
accept connections on ports you don't know how to handle. But then,
let's say that you want to make a new service in ANet. If it worked like
TCP (behavior is not defined), everyone on the network would have to
download your service module to make the module work on the network,
otherwise you might find it difficult to connect to the network, as no one
can accept the connection for your service, as no one knows how to handle
your "packets". The idea of splitting structure and behavior works great in
TCP/IP, as you connect to one specific IP address, but that cannot work in
a distributed network.

So, behavior has to be defined in ANet.


>If you want a chat type system then you'll also need - please transmit this
>data from here to this group of addresses.

That shouldn't be allowed. The goal of ANet is to have no need to be
identified by any "address" whatsoever. So, either you "find" the addresses
(or more precisely the proxies) that want the data with queries and then
upload the data to them, or you duplicate the data to the whole network as
static data.

Chat should then work as queries, as it already did some time ago in Gnutella.

- Benad