Why on earth do you have a 6148A-GE-TX (junk) blade in this switch at all? Can't you hear the switch running slower?
What is connected to that blade and what speeds are those devices using (10/100/1000)?
The bigger problem than the fabric switching-mode is the architecture of the 6148-GE-TX blade (Note that the 6548 has the same architecture on the port side, albeit CEF256):
Not sure you are aware, but that blade is configured as 6 groups of 8 ports. In other words, a 6 Gigabit blade, not 48 Gb (or 20 or 40). This is a wiring closet blade for *very* lightly used PCs connected at a Gig. NOT for a server farm, distribution, or core.
There are several problems with this blade:
1. There are not per-port buffers, only per-group buffers across 8 ports.
This can create buffering issues that *will* effect all 8 ports. The bad thing is, this can happen even if all the ports are running at 100 Mb!!!
2. Drop counters are not exposed to the cli because the 'hidden' pinnacle ASIC has no idea which port the buffer problem effects. This behavior can be modified by using the 'hol-blocking enable' hidden command on the ports, see next item. This effectively carves up the per-group buffer into smaller chunks and managing them per port in software.
3. Head of line blocking (HOL) is a *huge* issue with these blades. Basically a bunch of traffic can be *destined* for a port-group and starve the output buffers for all other ports in the group. This will essentially stop traffic to all other ports in that group until the HOL blocking issue dissipates, which could be never. This is particularly a problem with a heavily utilized server and/or using a port in the port-group as a span destination.
I have clients that have these blades who have been told by Cisco in TAC cases that they purchased the wrong (cheaper) blades and are using them for the wrong purpose and will need to use only 6 ports on the blade (every 8th port) to resolve their performance issue! Cisco documentation clearly states that these blades (6148-GE-TX, 6548-GE-TX) are *only* wiring closet blades.
Just my 2 cents since I see a lot of these conversations on these forums about the fabric switching-mode when the incorrectly used blades are a far bigger issue. Who cares how fast traffic gets switched between the blades when the 6148-GE-TX blade itself is 8 to 1 oversubscribed before the traffic even makes it to/from the backplane!!!
