Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!feed1.news.erols.com!insync!uunet!in3.uu.net!192.220.251.22!netnews.nwnet.net!Symiserver2.symantec.com!news From: tedm@agora.rdrop.com Newsgroups: comp.unix.bsd.misc,comp.unix.bsd.bsdi.misc,comp.unix.bsd.freebsd.misc,comp.unix.bsd.netbsd.misc Subject: Re: Running several networking cards in one system? Date: 13 Jan 1997 10:22:40 GMT Organization: Symantec Corp. Lines: 52 Message-ID: <5bd2dg$rkf@Symiserver2.symantec.com> References: <6OBfLaMbNgB@me-tech.pfm-mainz.de> <6OhJND_6NgB@me-tech.PFM-Mainz.de> Reply-To: tedm@agora.rdrop.com NNTP-Posting-Host: shiva2.central.com X-Newsreader: IBM NewsReader/2 v1.2.5 Xref: euryale.cc.adfa.oz.au comp.unix.bsd.misc:1887 comp.unix.bsd.bsdi.misc:5544 comp.unix.bsd.freebsd.misc:33805 comp.unix.bsd.netbsd.misc:5112 In <6OhJND_6NgB@me-tech.PFM-Mainz.de>, mschmidt@me-tech.PFM-Mainz.de (Michael Schmidt) writes: > In article <5b6eop$o0h@uriah.heep.sax.de> > j@uriah.heep.sax.de (J Wunsch) > wrote: > > >It may look pointless to you as it seems that you missed the point. In the >mentioned setup each NIC has to have a separate wire/cable (goes without If each NIC in the server has a separate wire/cable, what your talking about is good old fashioned routing. If these separate wires are all plugged into each other, there is no point under 10baseT to have multiple NIC's, since they are all one ethernet, and a single 10BaseT card can easily be driven by a Pentium to saturate the 10baseT segment. What is so complicated about all of this? All this load-balancing nonsense arose from the marketing departments of companies producing 100BaseT hubs, as a way of selling more hubs. There are a very few hubs out there that can take multiple connections from a server, and make it appear as though it is a single faster connection from the server to the hub. The drawbacks are that the hub has to be a switching hub, and you also have to run software on the server to do this. However, with these "load balancing with a switch" schemes, in essense they can be reduced to a network of separate ethernet segments, routed within the switch by mac address. This is not any different than implementing the same thing with standard hubs and multiple servers configured as routers, and just adding multiple NICs on different segments to the server you want to be well-connected. The thing that drives these "load balanced" schemes is the sucky IPX protocol, because by default IPX broadcasts are forwarded. This makes an administrative nightmare when you are attempting to build a network 'fabric" within a building, because of the nature of broadcast traffic as you continue to add network segments you can get into situations where you lose broadcasts from remote servers on a segment, then you run into clients connecting to servers through non-optimal routes. Thus, the IPX crowd hates to deal with multiple routers, they often don't understand IPX routing to begin with, and NetWare servers until recently offered insufficient control over IPX route advertisement to begin with. This, plus the use of the NetBEUI protocol has really driven the sale of expensive switching hubs intended to replace routed networks that were built improperly to begin with. If you use a decent protocol like TCP/IP, and interconnect your segments with each other through a network of servers routing each other, even a very simple routing protocol like RIP is often sufficient to use. In this case, you can approach the advantages of a switching hub network without the cost of a large switch. In this case, you simply add multiple NIC's to whatever server you want to be well connected, and spread your clients out over the network. You don't need any fancy load-balancing nonsense here, because your routing is taking the place of that.