David Rankin wrote:
>
> If someone else can enlighten me on how to have an "auto-sensing" hub handle
> both 10 and 100 Mips WinXX machines on the same lan segment, I am all ears.
The key in this statement is "hub"...
Simply put...
A hub is a "repeater" which implies all ports must be at the same speed; think
"bit copier". Usually these devices will have a small "elastic buffer" to
handle minor deviations in various sender speeds. If the sender and hub clocks
vary by more than the capacity of this elastic buffer (usually sized to match
deviation specs -- i.e., +|- 1KHz at 10Mb), the buffer will over|underrun and
the hub will jam the sender (forced collision). This will cause long packets to
fail while short ones succeed making troubleshooting "interesting". Hubs, being
repeaters, copy traffic to ALL ports. So 90Mb (100-10 or abs(10-100)) is
humongously out of spec for either speed.
A [multi-port] bridge ("[layer-2] switch" these days) buffers the entire packet
and forwards to the port where the target is known to be connected to (based on
MAC address), or all ports if unknown. Since the entire packet is in a buffer,
there is no longer a requirement to match speeds. Except for still allowing
mis-matched speeds, a switch can end up being no more effective than a hub
(sending to all ports) if the number of local hosts exceeds the size of the
address cache (used to select output port).
A router ("[layer-3] switch") also buffers the entire packet and forwards to a
port after the target's MAC address has been resolved (last router in path only;
intermediate routers forward based on routing tables). If you've ever noticed
the first ping packet being lost, that is due to the far-end router dropping the
packet, and issuing an ARP request to get the target's address. The router
expects the sender to retry. This situation doesn't occur very often, only if a
target has been idle for longer than the ARP timeout.
OK... more info than necessary. The point is that hubs are just dumb bit-level
repeaters requiring all ports to be at the same speed to work. Funny thing is,
they are no better than cable-only LANs, and may be worse if one or more senders
(hosts) have out of spec TX crystals triggering jams.
> (1) Set all 14 PCI WinXX machines and the Linux box to run at 10 mips (via the
>drivers); or
> (2) Find a 10/100 ISA NIC (3Com 515TX -- Costs more that the ISA machine is worth);
>or
> (3) Get rid of the machine that doesn't support a 10/100 NIC and replace it with one
>that does. (I don't care if it's nothing more than a 233 Mhz P2 with an open PCI slot
>for a 10/100 NIC)
(4) use a switch in place of the hub.
(5) use a cheap 100<-->10 bridge (or switch) between the slow box and the hub
(6) use a switch or router between different speed LANs
And don't put more than three hubs between any two devices on a LAN segment,
lest you want to learn about a whole new set of problems the hard way...
HTH,
Pierre
