[NTLUG:Discuss] Router or switch?

[Darin W. Smith]

On Mon, 05 May 2003 17:36:36 -0500, Joel Sinor <jsinor at comcast.net> wrote:

> traffic is sent to ports.  However, one of the things I have run into as
> a justification for higher-cost switches (in addition to the management
> and higher-level protocol handling you mention )  is bandwidth
> sharing/nonsharing. I still don't understand the specifics as much as I

You are correct about this.  Cheapo switches won't support 100Mbit going on 
all ports simultaneously.

Very large installations usually use a high-end switch that has one gigabit 
e-net port, which goes to the server that all individual machines out there 
will be talking to.  That really eliminates one bottleneck.

For SOHO use, though, you can't beat a 10 or 20 dollar full-duplex 100BT 
switch with AutoMDIX.

I have noticed that some switches appear to perform better than others, 
likely due to this very reason.  Switches in general, high capacity or not, 
are good for a LAN design, as they eliminate a lot of extraneous traffic.

FWIW, what you mention was one of the advantages of 100VG-AnyLAN, which was 
initially a competitor to 100BaseT for a 100Mbit ethernet standard.  
Unfortunately, it was an HP "club-members only" 'standard' and therefore 
got soundly trounced.  It had some great things about it though:
a) by utilizing all four pairs on a Cat3 cable, it could operate at 100Mbps 
on Cat3 "voice grade" wires (thus the "VG" in the name)
b) the standard provided that any switch, just by nature of supporting 
100VG, could support a dedicated 100Mbps connection between any two ports, 
while the others might fall back to 60Mbps.  Therefore, a bandwidth-hungry 
application like streaming video would be guaranteed its needed bandwidth & 
the rest of the ports had a guarantee on their maximum bandwidth.

As for Beowulf clusters (I've hoped to build one, but so far have not), you 
would indeed be best served by more expensive switches that can support the 
high bandwidth on all ports.

I would speculate though, that most people will never see more than 60Mbps 
on any 100BT line.  It has been my experience in the past (with 10BaseT, 
10Base2, and 100BaseT) that it is very rare to ever see greater than 60% of 
the theoretical maximum throughput on a line.  Part of this is due to the 
overhead of the ethernet framing, part of it is limitations in NICs, part 
of it is limitations in installations (rare is the installation that 
doesn't violate EIA/TIA 568A/B somewhere), part of it is probably noise 
getting on the lines somehow (whether due to cable-routing, or injected 
through a switch or NIC) causing retransmits.

-- 
D!
Darin W. Smith
AIM: JediGrover