A proposal ... not set in stone, but I do try to follow these rules. Last updated Mon Apr 8 15:31:28 CDT 1996 These are the requirements that *I* typically require for a new server. Other servers will have their own requirements, which may or may not be similar to mine. Note that piglet.cc.utexas.edu is no longer in a good position to route other IRC servers (i.e. to be a hub.) Therefore, we are not linking in any new servers, and therefore this entire document really doesn't mean anything ... but you may wish to consider it, as other servers may have similar requirements ... - A full T1 network connection or better is the bare minimum. Multiple T1's or a faster connection are typically the norm of new servers. - The IRC server's existance must be permitted by the administration of the machine and the network that it's sitting on. This kind of goes without saying ... - The persons responsible for running the server must be reasonably knowledgable about IRC and Unix. They should be willing and able to answer most user questions that they encounter regarding IRC. - The IRC server doesn't need a terribly fast machine, but it does need to be reasonably reliable and to not ever get totally bogged down. A Sparcstation 1+ with 32 megs of ram can make a decent IRC server if it's dedicated. If it's not dedicated, you'll need a faster machine, and more RAM. PC's typically don't make great IRC servers, especially when they're running Linux. Free/NetBSD typically works better for an IRC server, but even so, PC's aren't the platform of choice for an IRC server. - Running a server requires that the rest of the IRC network put a lot of trust in you. People who are known not to be trustworthy or with a history of not acting in the best interests of the `IRC net' will typically be denied server links. - New servers will be L: lined - meaning that it cannot link in other servers until such time as it is deemed that the server and it's administration are ready for such a load. - New servers should not be compiled in debug mode, and should consider enabling LOCAL_KILL_ONLY. - As far as bots go, the server can allow or disallow bots as it sees fit. But, certain rules should be followed - - They should not allow more than one bot per person. - The admins of the server should follow their own bot rules - if they ban all bots, they should not be running their own bots. - Not every domain can have it's own IRC server. Therefore, if you allow bots from the local domain, I feel you should also allow bots from your net-neighbors, even from competing ISP's, and they should be subject to the same rules as local bots. - Your policies should be enforced. Note that these are *my* views, and the admins of other servers may or may not agree. But following these rules will make me more inclined to consider linking your server ... Each server puts a certain load on the IRC network, so we need to make certain that all servers are in fact beneficial to the IRC network as a whole. Possibly the most important factor in determining if an IRC server is needed is the local userbase, how many local users are on IRC, typically averaged over a week. While bots are typically counted in these measurements, it's assumed that the majority of these users are in fact active humans. If the site runs large numbers of bots, these bots will be subtracted from the average, and may actually be seen as a reason to deny links to this site. Currently, recent (and old) usage stats are available via anonymous ftp from ftp.comco.com in the directory /pub/irc/stats Currently, the cutoff figure I use for average userbase needed for a new server is 30 clients for an educational institution, and 35 for other organizations, typically Internet Service Providers. These numbers aren't set in stone, but they are general guidelines. For purposes of this determination, I usually go by the average number of `non-duplicated' userids on IRC from this domain. I will occasionally link new server with (somewhat) lower numbers of users if some of the following are true : - they are in a particularly good network location (T3, multi-homed, etc.) - the administration appears to really know what they're doing, and really appears to be looking out for the benefit of the IRC network as a whole. This usually requires that I know them personally. - There are no IRC servers in their network-area. This does not mean I will link a server just because that state/city doesn't have a server, as networks typically don't care about state lines. Of course, if I do link a server with a low local userbase, I expect them to make up for it by handling a lot of remote clients ... If a site has certain problems, I will be reluctant to link a server there. Possible problems include : - clonebotters from that site. - just lots of bots in general, especially if their numbers approcaches/exceeds the number of humans, or if they are being run by the admins/operators of the new server. - the potential IRC admins or any of their `ircops' have a history of causing IRC `problems' (clonebotting, flooding, nick colliding, channel takeovers, harassment, etc.) - lax security. IRC servers typically attract crackers like a magnet. - Other established IRC servers that are very close network-wise, that are willing to handle your local clients. These numbers assume that your server will be willing to accept connections from other nearby (net-wise) sites, especially those without servers, even if these sites are competitors (in the case of ISP's.) Very few ISP's (right now, Netcom is the only one) have enough users to justify an IRC server just for them. If after a short period of operation, your server doesn't have an average, of, say 100 users on it at any given time, the server links will be reconsidered ... each server slows the network by a small amount, and if the server doesn't have enogh users to justify it's own existance, it should be removed. Same goes for servers that turn out to be unreliable, as they hurt the entire net. For an already existing server, I don't look so much at local userbase as I do at how many clients it handles, on the average. If it doesn't average at least 100 clients, then it probably should be removed. Basically, due to the way IRC works, we need a few big servers instead of a bunch of little ones. If we let every site that wants a server run one (lots of people just want to be ircops, and then lots of sites just want one for the `prestige' it may bring them, and lots of people want to run one so they have a place to put their bots) then IRC wouldn't work at all. Hopefully the next version of ircd will not have this restriction ... but currently we're stuck with this limitation, and we have to limit the number of servers on the net ... - Doug McLaren, dougmc@ccwf.cc.utexas.edu