| Nowadays people tend to put servers directly on the FDDI backbone
and clients hang off Ethernet ports on switches connected into the
backbone. It seems like you want to keep servers on Ethernets; in
that case, I would put each frequently-used server on a separate Ethernet,
and small client groups on separate Ethernets, interconnected across
an FDDI backbone. In other words, your "config 2" with 2 switches is
more along the right lines, except that 64 clients on a single Ethernet
may be too many. With 2 EF switches you get 12 Ethernet ports connected
into an FDDI so you should be able to create smaller groups.
Anil
|
| Configuration #1 COULD BE a better solution depending upon what versions
of NetWare the customer is running and traffic flow and server locations.
Note 2497.2 points out that Novell IPX is a Request-Response protocol
where EACH request must be explicitly acknowledged. This request-response
per packet sees a 2x delay when going through a store and forward bridge.
First the originator station sends the complete packet to the bridge, then
the bridge resends the complete packet to the destination. The destination
station then acknowledges the packet by a return packet to the bridge. After
the bridge receives the complete packet, it then transmits the acknowledgement
packet to the originator. Compare this to both server and client being on a
single Ethernet where the originator station sends a packet to the destination
and the destination station sends an acknowledgement back to the originator.
This is why many customers that insert a bridge into a Novell environment
typically complain about "slower performance". Most other protocols do not
have this problem since they can transmit complete trains of packets without
receiving an acknowledgement.
If the customer has implemented the new "packet burst" feature in Netware
(see Note #2497.3) which allows a train of packets to be sent before expecting
an acknowledgement, configuration #2 may be OK. But without this feature,
configuration #1 might be better especially if:
1) Servers are physically colocated with clients
2) Specific clients NORMALLY link to specific servers
3) A traffic analysis has been done so you know how to group the
top talkers together
Basically with this knowledge, you can establish multiple collision domains
where each domain is a server with its primary clients. The FDDI could be
used for servers that are "common" to all users but not a primary server.
And everything can be interconnected via the switch.
|