Attention: The Cisco Support Community site will be in read only mode on Dec14, 2017 from 12:01am PST to 11:30am for standard maintenance. Sorry for the inconvenience.
WS-X6548-GE-TX uses Centralized Bus Forwarding as opposed to CEF720 Bus Forwarding used on WS-X6748-GE-TX.
The CEF720 uses 2x20Gbps connection to the switch fabric, which on this case is the Sup Module (no need for a switch fabric module).
WS-X6548-GE-TX is only 1MB buffer per port while WS-X6748-GE-TX is 1.3MB buffer per port.
WS-X6548-GE-TX is a hybrid, it supports backplane connection in switch fabric or bus mode while WS-X6748-GE-TX can only support switch fabric.
Personally, I see WS-X6748-GE-TX, the module of choice if you have a Sup720 since it uses the ideal Bus Forwarding for this hardware.
I agree with everything Edison says execpt that it is 1Mb per 8 ports on the 6548. You really need to watch this as there is a large amount of oversubscription.
As Edison suggests go with the 6748.
6548 is 1MB per port.
Are we talking at cross purposes here :-)
In this doc ( Table 4 ) it states WS-X6548-GE-TX has 1mb buffer per 8 ports.
Conflicting documents, very interesting.
Table 2.1 from the link I posted says otherwise
Yes very interesting :-)
I missed that table from your original link.
I know there is an 8:1 oversubscription rate on the 6548-GE-TX which is why it is used primarily in the wiring closet.
I believe this oversubcription comes from a 1Gb uplink to the switch fabric for a group of 8 ports. So 8 ports running at 1 Gig could only have 1 Gbps to the switch fabric.
If for arguments sake we assume it is 1mb per 8 ports ( and im not trying to start an argument ! ) do you know if this 8:1 buffer ratio is in any way related to the 8Gbps to 1Gbps switch fabric connection ?
Could be. Your link provided more detail than mine, so I'll go with the 1MB per 8 ports until I have some time and test it :)
it is already 1MB for 8 ports. You can simple test it. Connect two PC on port GeX/0 and GeX/1 and try copy data to another two PCes. You can check "total output drops" on the interfaces (show interface GeX/0 and show interface GeX/1). The "total output drops" value will be the same for both ports and will increment... We have many problems with port buffers sharing (for example in case that you mirror 10gigs to the 1 gig port via span sessions) -> other ports in 8ports group was dead ;o(
What are they referring to by 'wiring closet'. Does it mean what you connect your workstations & laptops to? If so, why would you have a 6500 in your IDF closets?
How does 1mb buffer for 8 ports translate to 8:1 oversubscription? Aren't the two unrelated?
Does anybody know what's the oversubscription for the 6748 ethernet blade if any? I believe it has a 2 X 20gbps links to the backplane giving it 40gbps total.
"Wiring closet" is a common term for where the user hosts connect to a network device. Often also the "closet" for phone wiring.
Whenever you have more bandwidth in then out, you usually see some amount of buffering RAM. The more you have, the less likely to drop packets/frames during short term congestion. (Of course, conversely increases latency, which is why the better devices offer CoS/QoS features.)
Depending on the hardware, actual buffer amounts can vary for the same oversubscription ratios, and where it actually has the buffers, ingress, egress or both.
"If so, why would you have a 6500 in your IDF closets?"
Many of the same reasons you would use it else where, premiere software, redundant sup and power options, wide choice of line cards (including WAN), service module cards, high port density (6513).
A user facing closet is also where you're more likely to deploy a 6500 with sup32 or sup32-PISA. (The sup32-PISA is especially interesting for the user edge.)