P2P-Zone - View Single Post

napho · 23-03-02, 06:51 AM

Great program but too slow now- this from the creator

Swamp
Donkey Master

Joined: Feb 15, 2001
Posts: 622 Posted: 2002-03-17 16:45
--------------------------------------------------------------------------------
Ok here is the first state of the network address:

Unfortunately what is happening now is the network is starting to reach its maximum load of users. I had envisioned the network growing in a slightly different way. I had assumed that there would be more isolated pockets of servers and that some servers would be able to handle far more users than about the 5,000 max now. So a temporary solution to the current problems are doing what I suggested here: http://forums.edonkey2000.com/phpBB...c=27814&forum=1

The reason this is important is the following:
Servers get UDP traffic and search requests from every users that know about it.
Each server can only handle X amount of UDP traffic and search requests before it is brought to its knees.
This means that each server can only have so many users that know about that server.
Right now every user tries to know about every server meaning that servers must handle the load of the entire user base something that clearly can't scale very far.
When servers are overloaded not only do they not handle extended search requests but more importantly they don't handle callback requests.
Callback requests happen when a particular user can't accept incoming connections (firewalled) (low ID).
When A wants to talk to B and B is firewalled, A must ask B's server to have B make the connection to A.
So basically when B's server is overloaded no one will be able to talk to B.
This is a big issue since it seems about half the users are firewalled.

I will probably release a few changes to the server soon that will facilitate making subnetworks and handling all the excess UDP traffic.

But the main thing I'm working on now is revamping the entire search protocol. The new protocol will eliminate the need for servers. It should also scale to tens of millions of simultaneous users. Unlike the other fully distributed approaches in use now, you will have access to every connected user's files.

So anyway we do hear your complaints and we are definitely working as fast as we can to fix the problems.
Thanks for hanging in there.

23-03-02, 06:51 AM	#1
napho Dawn's private genie Join Date: May 2001 Location: the Canadian wasteland Posts: 4,461	eDonkey to make improvements Great program but too slow now- this from the creator Swamp Donkey Master Joined: Feb 15, 2001 Posts: 622 Posted: 2002-03-17 16:45 -------------------------------------------------------------------------------- Ok here is the first state of the network address: Unfortunately what is happening now is the network is starting to reach its maximum load of users. I had envisioned the network growing in a slightly different way. I had assumed that there would be more isolated pockets of servers and that some servers would be able to handle far more users than about the 5,000 max now. So a temporary solution to the current problems are doing what I suggested here: http://forums.edonkey2000.com/phpBB...c=27814&forum=1 The reason this is important is the following: Servers get UDP traffic and search requests from every users that know about it. Each server can only handle X amount of UDP traffic and search requests before it is brought to its knees. This means that each server can only have so many users that know about that server. Right now every user tries to know about every server meaning that servers must handle the load of the entire user base something that clearly can't scale very far. When servers are overloaded not only do they not handle extended search requests but more importantly they don't handle callback requests. Callback requests happen when a particular user can't accept incoming connections (firewalled) (low ID). When A wants to talk to B and B is firewalled, A must ask B's server to have B make the connection to A. So basically when B's server is overloaded no one will be able to talk to B. This is a big issue since it seems about half the users are firewalled. I will probably release a few changes to the server soon that will facilitate making subnetworks and handling all the excess UDP traffic. But the main thing I'm working on now is revamping the entire search protocol. The new protocol will eliminate the need for servers. It should also scale to tens of millions of simultaneous users. Unlike the other fully distributed approaches in use now, you will have access to every connected user's files. So anyway we do hear your complaints and we are definitely working as fast as we can to fix the problems. Thanks for hanging in there.