You folks out there with large SANs.

Is anyone trying to do a good job of localizing and keeping it
maintained so that for a given resource(disk system or tape library)
everyone who uses it is on the same switch(s) the resource is on.
Assuming all 32 port switches and a disk system with 2 FC connections
plus a tape library with 2 FC connections and 2 isl's(1 into each core
switch and hopefully not used much) you're left with 26 ports for nodes
to attach.
On a second fabric you could do the same so you could have a total of 26
dually connected nodes and 1 disk system and 1 tape library. 

Is it even possible to maintain good localizing or is it better to make
your core big and edge cheap.
Currently we have a core/edge design with 2 core switches 16 port 3800
Brocades and 3800 edge switches with each edge switch dually connected
into each core switch.  So with 16 port cores and each edge dually
connected into both cores(4 isl's) we have 8 edge switches with each
having 12 useable ports resulting in 96 edge ports available.

Note that every edge switch is dually connected into both core switches
leaving only 12 ports useable for edge nodes.
Assuming we have budget only to buy two 3900 Brocades(32 port 2G
switch).

The dilemma we have is whether to: (ignoring for the moment any
potential over-subscription problems.)
        1) Use them in the core replacing the 16 port 3800's with 32
port 3900's giving us an immediate gain of 24 useable edge ports.
        OR
        2) replacing 2 16 port edge switches with 32 port edge switches
(and shelving for now the replaced edge switches) giving us an immediate
gain of (28+28) - (12+12)  or 32 useable edge ports.

Option 1 means moving the 2 old 16 port cores out to the edge giving us
24 more edge ports(12 on two switches).
        This also gives us 28 more free core ports where key resources
could be attached guaranteeing only 1 isl between a resource and any
port on the edge.(or later more edge switches could be purchased and
attached or ultimately we could have 16 edge switches and they could all
be 32 port switches giving us a total of 16*28 or 448 edge ports.

Option 2 means replacing 2 existing edge switches(16 port 3800's) with 2
32 port 3900's and shelving the replaced switches.  This would give us a
net increase of (28*2) - (12*2) or 32 useable edge ports.  And it would
be important to localize and maintain it.

Regarding potential over-subscription we currently have 12:4 on all the
edge switches.
Going with 32 port edge switches and not increasing the isl's would put
us at 28:4 which is probably not good unless we could nail the
localization and believe it can be maintained.

The bottom line is if we think SAN disk and nodes using it's LUNs will
continue to grow fast maintaining localization may not be realistic.  If
not we'd be better off expanding the core first.

And there's still the question of over-subscription.  To maintain our
current over-subscription of 12:4 we have now would require us to have
each edge switch quadruply connected into each core so that with 32 port
cores and 32 port edge switches we could only support a max of 8 edge
switches with only 24 ports available for edge use (since 8 would have
been used as isl's (4 into each core).  So we'd have 8*24 or 192 useable
edge ports which represents a doubling from the max of 96 edge ports we
have now.  But we'd need to start with replacing the 16 port cores.

Let me know what you think.

Thanks - John

--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=