Well, now you're really talking about the dark ages. ;-) You are back to the early 1990s discussion about upgrading hubs to switches. That's a good idea so that each port has 100 Mbps (or 10 Mbps) rather than all ports sharing bandwidth and being in the same collision domain. I can't think of any reason not to upgrade to switches at this point. It's difficult to even buy a 100-Mbps hub any more. (I tried and they sent me a switch!) The upgrade is quite seamless (unlike the upgrade from switches to VLAN-aware switches.)
This has nothing to do with the late 1990s question of broadcasts which came about when people started replacing routers with switches and designing a network that was a large broadcast domain. They thought they had solved all their problems but they hadn't because a switch forwards broadcasts, whereas a router does not, of course. VLANs let you divide up those broadcast domains and be smarter about the flooding of unknown unicasts (as someone else mentioned, which was a good point.) But VLANs bring with them all sorts of other management headaches. It's a tradeoff that doesn't need to be made in many modern networks, despite what Cisco tells you. The materials that we read about broadcasts in switched networks come from studies Cisco did in 1994. And some books still have that silly triangle that a Cisco marketing engineer (now that's an oxymoron!) designed in 1994. Yes, I know that VLANs have other advantages (supposedly) besides dividing up broadcast domains, and I warned people up front that my point of view was controversial, but I'm sticking to it. ;-) With regards to your practical limits, Cisco has some guidelines (but once again they are based on OLD data ;-) A broadcast domain shouldn't have more than a few hundred nodes. Also, with regards to your comment about sniffing on a switched network. Remember that all you see is broadcasts and traffic to your port (unless you mirror other ports) so you get a skewed view. So have we beat this one to death yet? I enjoyed the discussion. (I hope we didn't put everyone else to sleep! ;-) Priscilla >Well, I admit, my response was a bit clouded by the fact that one of our >clients recently requested a redesign of their flat beyond flat >network. Call it justification! They are using, UGH, 10BaseT Hubs with >some nasTY (with an iintentional capital T and Y), daisy chaining hub >action, which REALLY exacerbated performance loss. Not to mention it's >all Bay GEAR! Evil! :) Admittedly, that IS changing the premise of >Priscilla's original statement. The network I am working on is HARDLY the >epitome of the modern day model system Priscilla described. I am guessing >with solid switches across the board, it might very well be "pretty darn >good" in terms of performance. I was just curious where the new practical >bar was raised to. > >If the situation is with 10BaseT hubs, I would not be surprised if >performance is really becoming an issue where broadcasts become a >percentage of your daily bandwidth. Where broadcasts are probably far more >often being that even unicast packets are broadcasted on the wonderous >layer 1 repeater technology known as hubs. With all switches, I am not too >sure I can say clearly otherwise, but I was just wondering "how far" is a >practical limit in today's modern systems? On top of that, yes, all in >moderation. If we take either approach to the extreme, we clearly see >significant flaws. No one wants to run subnets of 2 usable hosts each for >their entire network and smash their catalyst 6509 with routing modules to >oblivion. No one wants to run the 30,000 flat network from HecK. (Ok, >maybe some people do...) "Look Ma, no routers!" > >On the side, you just noticed your statement impies that some would run >multiple VLANs with a single subnet? I guess you would depend on having >at least one port on both VLANs to get interconnectivity? Would that be >like bridging? (unifying two layer 2 networks). > >Her statements on the windows protocol seem correct. Ugh, I got to whip >out the old sniffer again. Or read up again. I could have sworn I STILL >saw a multitude of crap flying every second on my old college network even >after we went to a switch. I should try again since her points seem quite >valid. > >Hm. Although broadcasting was necessary, in the more extreme case, does it >make sense for a quote server to broadcast to another quote server? There >is a small subsegment of "don't cares" for the quotes, it seems like >multicast is more ideal, but probably not necessary. No matter, I am sure >the demigods of broadcast control had a working solution. :) > > >-Carroll Kong ________________________ Priscilla Oppenheimer http://www.priscilla.com Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=24124&t=23950 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
