home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: alt.irc
- Path: sparky!uunet!decwrl!lambda.msfc.nasa.gov!sol.ctr.columbia.edu!mtu.edu!ngmitche
- From: ngmitche@mtu.edu (SuperNate I am!)
- Subject: Re: How many clients to a server?
- Organization: Michigan Technological University
- References: <1992Jul20.085423.25491@aston.ac.uk>
- Message-ID: <1992Jul30.023524.19119@ctr.columbia.edu>
- Distribution: planetary
- Sender: news@ctr.columbia.edu (The Daily Lose)
- Date: Thu, 30 Jul 1992 02:35:24 GMT
- X-Posted-From: melab9.me.mtu.edu
- X-Posted-Through: sol.ctr.columbia.edu
- Lines: 59
-
- The general concensus of the discussion about servers seems to indicate that
- a server will use at least as much bandwidth as 10 clients, and since the
- server runs continually, will do so at all times.
-
- I do not have a functioning server (although a fair number of people have
- suggested that I make one), so I have no confirmed information to amend
- this approximation, but would like to point out a few things.
-
- An unloaded server might use the same bandwidth at most times of day as 10
- regular clients, but since the load caused by a server increases as clients
- are added to it, the "break-even" point is probably closer to 12 when
- a simple comparison of individual loads is made.
-
- That is: the loads (bytes, packets, whatever) of 12 clients is compared to
- a comperable measurement of the load of a single server. This provides a
- simple comparison, but still neglects certain facts. In general, persons from
- a local region will converse more with each other than with other users.
- Also, such things as notify and channel modes might also be similar. So for
- those messages that are sent to several clients from a local region, it might
- be more efficient to have a local server so that they will cross a long stretch
- of the net only one time. This will once again change the "break-even" point
- in the opposite direction, but probably not below 8. (8 clients == 1 server)
-
- The chief concern of any site running a server is the load on that site's link
- at PEAK hours. Since the load on the site will almost NEVER be less than
- that of 8 clients unless there are VERY few people on all of IRC a simple
- suggestion is just don't run IRC at peak times of network use. Sure it's not
- as nice as getting a server running and having savings from going to 20 clients
- to the equivalent of 10, but getting the number of clients down to less than
- 10 during peak hours will show net savings without the hassle, work, and off-
- peak constant load of a server. No IRC between 1400-1700 for instance.
-
- Peak IRC hours aren't usually at the same time as peak network hours (thank
- god) so it might not REALLY be a killer problem. That is, the server
- won't be bogged by keeping track of a zillion users at the same time that
- everyone and their brother at a site is ftping, emailing, (even ircing),
- telnetting, etc.
-
- If I don't see at LEAST 8 clients during peak hours, I don't need to make a
- server and since MTU has a policy disallowing IRC at any time that might be
- construed as "high-network-usage", I sure hope that I don't EVER see that
- many IRCers from tech at those times. I would like to continue to use IRC
- and not have it squelched because a few individuals cannot respect the limits
- of the resources available to students here. I can think of one person who
- ran IRC when his department's mainframe was down to 4M and swapping to a floppy-
- drive or some rediculous situation like that--and YES, it was during a time
- that MTU policy would ordinarily disallow IRC. What bugs me most is that _I_
- set up the client on that mainframe and before I did so, I specifically asked
- that user to talk with the director of that department's computing facility so
- that he would be fully aware and cognizant of the policy. (yeah, right!)
-
- All that I have mentioned is just stuff to consider. I'm not a computer whiz
- or a networking whiz. I've only used real computers a little less than a year,
- but at least my default server list for the client I maintain uses servers
- that are topologically close, reliable, capable and have consistently small
- ping times..
-
- --
- What? no .sig?
-