| Title: | DEChub/HUBwatch/PROBEwatch CONFERENCE |
| Notice: | Firmware -2, Doc -3, Power -4, HW kits -5, firm load -6&7 |
| Moderator: | NETCAD::COLELLA DT |
| Created: | Wed Nov 13 1991 |
| Last Modified: | Fri Jun 06 1997 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 4455 |
| Total number of notes: | 16761 |
We're responding to an RFP - college campus...
Requirements:
FDDI backbone connecting 2 large Ethernet switches with DAS ports.
Ethernet switches will provide ports (fiber) for remote building LANS,
max. 60 nodes per LAN.
Total switch ports called for - 75, split is roughly 50 for 1 switch
and 25 for the other.
Remote buildings call for 10baseT connectivity via stackables or
chassis-based hubs.
I'm looking for creative solution using Digital products. The best idea
I've come up with so far is to push the switching function further
out into the network. For example, at those points in the fiber
distribution where 6 LANs come together, place a DEChub 900 with a
DEFMM and a 900MX bridge and create a FDDI backbone of about 15
switches rather than just 2. This strays from the spec a bit but... I'm
open to suggestions. ...15 900MX bridges back to a GigaSwitch?
Competition includes all the major players - Chipcom with their
Galactica switch, Cabletron, Synoptics, and who knows who else.
Net mgmt is important as well, platforms must be UNIX, MAC, or DOS.
BTW, they are asking for an Off-Campus Network Access Plan... any
pointers? And a network security plan too. Again, any pointers?
| T.R | Title | User | Personal Name | Date | Lines |
|---|---|---|---|---|---|
| 1344.1 | When will it BREAK? | NCMAIL::ADAIRC | Mon Sep 19 1994 16:43 | 19 | |
Having submitted a proposal with a distributed switch approach the
question comes back from the customer regarding loading on the FDDI
ring. We have 15 switches sitting on the ring (DAS all the way around)
each serving 6 Ethernets. That's 90 Ethernet segments being bridged into
the FDDI ring. (A maximum of 60 nodes per Ethernet segment)
Traffic characterization information is limited. This is a new network
and the existing network does not come close to approximating the new
network.
So, is there some technical performance/loading info regarding FDDI
network performance that would at least provide a framework for this
customer to think intelligently about his question?
Has anyone done any modeling on a configuration like this?
Suggestions?
Claude
| |||||
| 1344.2 | NETCAD::ANIL | Tue Sep 20 1994 00:09 | 20 | ||
A shared Ethernet LAN designed conservatively would have
an average traffic of about 30%, with peaks to 50%. With
90 Ethernets, you get a total of 90*10*0.3 = 270 Mbps average
and peaks of 450 Mbps -- which clearly exceeds a single
FDDI's max bandwidth of 100Mbps.
Now all the traffic does not have to end up going through the
backbone. A workgroup-based network design would use the backbone only
when necessary, to talk OUTSIDE of the workgroup - that is, a
a significant percentage of traffic will be exchanged between the
Ethernet ports of a DECSwitch 900EF which will not go across
the FDDI. So the numbers above can be downgraded by that amount.
If you believe that the FDDI will carry the brunt of the traffic, you
can do one of two things: use multiple FDDIs interconnected by
back-to-back DECswitches through Ethernets. Or, you could use
a GIGAswitch to connect the FDDI's directly (although it appears
to be too late for you to do this).
Anil
| |||||
| 1344.3 | NETCAD::SLAWRENCE | Thu Sep 22 1994 15:22 | 16 | ||
I suspect that Anil's assumed ethernet loads may be rather pessimistic
(high).
However, the real issue is how the traffic will be distributed; if each
of those local nets is a few servers and a lot of workstations - and
those workstations will be mostly using the local servers, then I
suspect that the load going out onto the FDDI will be easily low enough
to make this work. If on the other hand all the nodes on the ethernets
are diskless workstations mounting all disks from servers located on
the other side of the FDDI, then you'll have real problem.
The latter would not be good distributed system design no matter how
fast the backbone is, since you can never scale the backbone speeds as
quickly as the end nodes can increase the load (there are always too
many of them :-).
| |||||