NewNet


Working Toward a Fully-Meshed Network Topology
Written by nanook on or about April 19, 1996

Existing EF-Net Topology In the existing EF-Net topology network connectivity is unreliable and abuse of power is rampant. The reason that network reliability is poor and that splits are common is that the failure of a single HUB site or link to that HUB site can isolate portions of the net. This also creates a situation where power is concentrated in those central HUB sites and that power tends to be abused. Those who presently control the central HUBs in EF-Net have resisted any attempt to make the network more fully meshed because they realize this would diminish their power over the net.

The probelms that exist in EF-Net today are problems that are inherent in a star topology. It also introduces severe scaling problems as some central HUB will need to pass the majority of the traffic and the capacity of that central HUB ultimately sets the upper limit on the size of the network. A more fully meshed topology will eliminate this bottleneck and allow the net to grow. The next major hurdle limiting network size is maintaining state information for the network. This also is a major problem for the Internet as a whole.

Matthew Ramsey and Jonah Barron Yokubaitis were upset about our bots here adding to the number of network connections and because I refused to disallow bots on our server they cut our server, IRC.ESKIMO.COM, off of EF-Net.

Why would this be such a big issue to them? Because they operate these HUB sites that are nearing or over capacity. The fact that they would be that concerned over less than 1% of the total network connections indicates their servers must have been damn close to saturation if not totally saturated. A more fully-meshed topology will distribute the load more evenly across servers and network links allowing the network to scale better.

More Meshed Topology The fact that they operated central HUB servers also gave them the power to cut us off. As far as I am concerned, this is another major design flaw in EF-Net. Other IRC OPs for the most part go along with what these folks dictate because the result of opposing them is what happened to us, banishment from the net.

To remedy the network reliability issues and avoid the concentrations of power that have occurred in EF-Net, I am advocating a more fully meshed network topology. In this topology EVERY server would have several connections to other servers and the loss of any HUB or network connection would not isolate any portion of the network.

It also provides greater scalability. This is important to the future of IRC. Right now IRC is almost entirely text. But I forsee a day when people are going to want to do new cooler things, video conferencing, audio, etc. The XDCC mechanism is not adequate to this task except for one-on-one communications. The IRC server network as it exists in EF-Net would not sustain the bandwidth requirements of multi-media applications. Again, a more fully meshed network in which the distribution traffic is spread out over multiple servers is the answer.


[  Info  |   News  |   People  |   Contribute  ]

[  Servers  |   Services  |   Policies  |   Support  |   Chat  ]