[MUD-Dev] How to support 1000+ simultaneous connections, and some philosophy.

J C Lawrence claw at varesearch.com
Thu Mar 11 18:32:27 CET 1999


On Thu, 11 Mar 1999 08:03:09 -0700 
Chris Gray <cg at ami-cg.GraySage.Edmonton.AB.CA> wrote:

> [Oliver Jowett:]
>> On Wed, 10 Mar 1999, Chris Gray wrote:

>> What about using round-robin DNS (multiple A records for the same
>> hostname)? It's using the same port, just on different IPs that
>> all happen to be on the same machine.

> I'll let others answer that - my knowledge of networking is mostly
> as a programmer, and not as an administrator/configurator. I'm not
> sure how easy it would be to do (not in editing files, but in
> practical matters), or whether it would yield decent balance among
> the 10.

To very quickly recap and illustrate the area:

  It is possible to configure a fully qualified domain name so that
it maps to multiple IP addresses.  This is done very simply by just
defining multiple IP addresses for that domain.  

  Want an example?  Run `nslookup` against yahoo:

    $ nslookup www.yahoo.com
    Server:  ns1.varesearch.com
    Address:  209.81.8.8

    Name:    www.yahoo.com
    Addresses: 204.71.200.74, 204.71.177.71, 204.71.200.66,
    204.71.200.72, 204.71.200.69, 204.71.177.70

  The end result of doing this is that the name server cycles among
the IP addresses when answering resolution requests.  Thus, the
first client asking where "www.yahoo.com" is gets, "204.71.200.74",
the next client gets "204.71.177.71", the next, "204.71.200.66", and
so forth until the end when it starts all over again.

  This is a very crude form of load balancing and is only really
effective when your user transactions are both short and of
reasonably uniform size/expense (eg web hits).  If your transactions
are vary significantly in length, there is the possibility of
(randomly) all your long/expensive transactions getting assigned to
one node, while all the quickies get assigned to the rest, leaving
one node drowning while the rest cruise.

  More intelligent DNS load balancers maintain an arbitration system
with the nodes they represent, and assign new tasks to least-loaded
nodes.  There are also many other forms of load balancing, encluding
supports that actively migrate processes across nodes to ensure aven
loading (IRIX does a bit of this in their CC:NUMA machines).


--
J C Lawrence                              Internet: claw at kanga.nu
---------(*)                        Internet: claw at varesearch.com
...Honorary Member of Clan McFud -- Teamer's Avenging Monolith...


_______________________________________________
MUD-Dev maillist  -  MUD-Dev at kanga.nu
http://www.kanga.nu/lists/listinfo/mud-dev




More information about the mud-dev-archive mailing list