Hi, James.

I suspect you know much (if not all) of my comments below, but I'm going
into extra detail for the edification of others.  [IMO, Raven's archives
have to be one of the more critical bits of infrastructure to preserve for
MPE homesteaders and hobbyists going forward.]

At +0000 05:47 AM 3/26/2003, James Hofmeister wrote:
>Has any one had experience with the hp ProCurve Switch 2708 ?
>
>http://www.smb.compaq.com/optionCategories.asp?oi=E9CED&BEID=19701&SBLID=&SuperCategoryId=168
><watch the wrap>

http://tinyurl.com/86cg

>a low-cost, unmanaged 8-port switch with 10/100/1000 auto-sensing per port
>and hp Auto-MDIX on all 10/100/1000 ports

$690.34

Once again "low-cost" appears to be one of those phrases of matchless
elasticity.  :)

>I have a diagnostic h/w room with a mixture of 10Mb, 100BT and now a few
>Unix 1000BT Gigabit network links.
>
>Most of the 10Mb system links (40+) are randomly connected to several of
>the hp 8-port 10Mb hubs;  The 100BT system links (5) are connected to a hp
>8-port 100BT hub;  We have 3 new systems capable of 1000BT Gigabit Ethernet.

I presume you (and the HP website) are using the terms "hub" to mean a
device which blindly broadcasts all incoming messages to all ports (at the
same speed) and "switch" to mean a device (sometimes called an "N-way
switch") which pays attention to what Ethernet addresses are on which port
and optimizes traffic by sending messages to known devices out only the
port where it has seen the device previously (but all unknown addresses are
still broadcast).  Note that a dual (or more) speed hub (e.g. 10/100 hub)
is in reality a dedicated switch connecting logical hubs (auto-partitioning
in the case of auto-speed-sensing ports) that run at a single speed.

For example, in the case of a 10/100 hub, the 10Mb devices are effectively
on a single 10Mb hub (and hear all of each others traffic) and likewise the
100Mb devices are effectively on a *separate* 100Mb hub but where the two
hubs are connected by an internal switch.  [It's essentially required to
put something like a switch between two different speed hubs like this
because otherwise the traffic from the faster hub could overrun the slower
hub.]

Hubs and switches can be viewed as the LAN equivalents to bridges and
routers (respectively) in a WAN.

>I would like to connect the 3 1000BT systems each to a port in this
>switch, the 1 100BT hub to a port on this switch and the several 10Mb hubs
>to the remaining ports on this switch saving 1 port as a path back to a
>10Mb backbone...

That topology leaves open the maximum number of high-speed ports for future
use (which may be optimal for a low-bandwidth diagnostic lab).  However it
is not optimized for maximum potential throughput.  Having all the 10Mb and
100Mb devices on the 10/100 hub means that all the 100Mb traffic has to
share a single 100Mb pipe (both to talk to the 1000Mb switch/systems and to
talk to each other).

If you move your highest traffic load (presumably but not necessarily
100Mb) nodes to the 10/100/1000 switch then the switch will have to work
harder but those devices can all have a dedicated channel to talk to each
other without having to share bandwidth that's not relevant to them.  And
the 100Mb pipe to the 10/100 hub will only have to carry traffic to/from
the 100/MB devices (plus any traffic to/from the 10Mb devices to any of the
100Mb or 1000Mb devices.

>The utilization and travel levels on this network are extremely low.

Then the above probably doesn't matter.  Plug them together however feels
good.  :)

>I am also considering only handling the 100BT and 1000BT traffic with this
>switch again saving 1 port as a path back to a 10Mb backbone...

Just remember that each of the ports on the switch outputs only traffic
that needs to be on that port, but incoming data (like from a hub) will
have the aggregate of all the traffic on that hub (at least at the same
speed as that port on the hub).

I generally visualize a dual/multi-speed hub as separate hubs with a switch
between each of them, and an N-way switch as N "single-port" hubs connected
by a switch.  Maximal bandwidth will happen when you minimize the traffic
at the busier hubs and cable segments.

>Any feedback ?

FWIW.

>Thanks ahead of time.

You're welcome.  :)

--
Jeff Woods <[log in to unmask]>

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *