wrote in message [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > > > Hi all, > > > > > > I can't for the life of me remember what the recommended > > > maximum number of clients on one segment is, I think it was > > > either 300 or 500. > > > > > IP is 500 > > IPX is 300 > > AppleTalk is 200 > > NetBIOS is 200 > > Mixed is 200 > > > > > > Also, those numbers from Cisco are pretty dated. These days switches with > > full-duplex ports are so cheap, you can have most of your LAN "segments" > > with just two nodes on them! (The PC and the switch port) > > > > > > Priscilla > > Hi Priscilla, > > Lol, now I've got all confused :-) > > Do these figures relate to broadcast domain or collision domains ? I've > always assumed they were broadcast domain. > > Also, would the speed of the wire make a difference ? > > I'm asking 'cos I've just joined a new company. There's 700 devices using > TCP/IP and NetBIOS on a flat network. Everything is connected via 100BaseT, > mostly one device per switch port and all switches are connected via fibre. > Ping packets (indeed all types of packets) are just "disappearing" as are > mapped drives. Machines crash out, etc, etc. > > My starting point was the obvious, subnetting the network. I am miles out ? > Any gentle push in the right direction would be greatly appreciated :-) > > Many thanks, > > Graham.
I manage a ~1,000 node network. All of the computers (>800), to include servers, are NT 4.0 and all run NetBIOS over TCP/IP, only. The network is completely flat, but everything is at least 100Mb FD. Everything is working fine. As others have said, the traffic really needs to be analyzed and performance evaluated to determine what is 'too much'. I've spent some time with sniffers and MRTG to find "bandwidth hogs", actually with more emphasis on multicast/broadcast hogs. The browser service has been disabled on all workstations (can cause huge spikes during browser elections), WINS has been installed, all unnecessary protocols have been removed, etc. The average non-unicast rate is about 15 packets-per-second, which is about 1% of 1% of the available bandwidth. Processor utilization on a 400MHz machine doesn't rise above 1% even during non-unicasts going over 100 per second. One of the reasons for the low CPU hit is because most of the non-unicast packets aren't for the NT boxes and the NIC cards don't even let them in. Packets such as BPDU's don't usually effect a computer. Now here's the amazing part (at least I think so). When I got interviewed for the job, my boss (he is now) showed me the network diagram. At the time EVERY computer was running TCP/IP, IPX (to include CSNW) and NetBEUI. Some servers were even running DLC as well. There was only switch which split the backbone for all of the workstations (at the time it was >500) into only 12 10Mb collision domains. Yes, collision domains, not broadcast domains. All of the closet boxes were 10Mb hubs. I looked up at him and said, "This actually works?" I think that is one of the reasons I got the job, especially since my predecessor left a couple of months before. Anyway, what is amazing is that even that scenario 'functioned' but it definitely needed 'tuning'. Once again, I can't empasize enough that you have to know how busy your network is, identify what isn't needed, remove what you can, tweak what's left (if possible ), etc. It is unfortunate that none of the popular certifications emphasise these skills, which data plumbers really should have. As Confucious said; "One must not only know how to configure an interface, but also know what goes through that interface." OK, he didn't really say that, but maybe he should have. ; ) HTH, Clay Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=49844&t=49704 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
