Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
CEF720 will use the shared bus for lookup http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps708/prod_white_paper0900aecd80673385.html
Agreed (see Figure 24 in your reference [also for CEF256]), although dCEF should not (see Figure 25 in your reference).
Yes, 6724 as well as the 6748 will make use of an optional DFC and they become dCEF720 modules once this option is added increasing their performance.
Usage of Fabric does not require a DFC but the module needs to perform lookup using the share bus hence there is a minimal performance impact.
". . . modules needs to perform lookup using the share bus . . ."? You mean a CEF line card or the DFC module? Agreed if you mean the former, if the latter, DFC should be doing lookup locally.
I also agree using the main sup vs. a DFC for the lookup may incur some performance impact (probably increased latency, but if so, probably also at the microsecond level), but assuming the main sup isn't overload, it's probably very, very minimal impact. In other words, just using DFC enabled CEF line cards might offer unnoticeable improved performance unless the main sup is overloaded.
However, I made sure to use a module such as 6708 where the DFC isn't optional to make my point.
As an aside, the 6708 is an interesting card, as you note, it comes standard with a DFC. Only Cisco could really tell us why they made the DFC standard with the 6708, but my guess is 1) because the card oversubscribes the fabric (2:1), believe the DFC allows same line card port-to-port traffic, which avoids the potential bottleneck to the main fabric, and 2) when the card card came out, you were most likely going to use it for infrastructure links, not host links, and the former might be "busier". If so, and if you start to load up a core/distribution 6509 with these, the overall aggregate bandwidth could easily outpace the capacity of the sup720, so off-loading packet forwarding to the line card may become critical. (NB: BTW, in theory, the 48 Mpps offered by the DFC4 only guarantees about 32 Gbps [i.e. for min size Ethernet].)
PS:
To the original poster, if you mix fabric only cards with classic bus cards in the same chassis (e.g. your 6148 and your 6724), the cards might not directly transfer data between them (fabric only cards put their data on the fabric, which a classic bus card cannot "see"; unsure whether fabric only card will "see" data on the main bus - data can be transferred between main bus and fabric on sup720) and you're unable to use "compact mode" (reducing the main sups maximum forwarding performance by half or more).
If you mix classic bus and CEF256 cards in the same chassis, I recall there's an option to tell the CEF256 cards whether to place their data on the share bus or the fabric. The former enhances data transfer between a classic bus card and a CEF256 card, the latter enhances data transfers between fabric cards.
Basically, you can mix and match many line cards in a 6500 chassis, but some line card mixes work better together. Overall performance for "same" port speed, on different cards, can vary due to architectural differences.
Unless you have very busy line cards (or high aggregate across all the line cards), DFCs might be one of the last things you need to worry about. I've seen Enterprise 6500s work fine loaded up with with gig and 10g ports, but all just running on the main sup, and I've seen Service Provider 7609s maxing out DFCs on 6704s.
On older line cards, e.g. again like your 6148, it's very common to see port drops when running its ports at gig. This can be somewhat mitigated by how you distribute your hosts to port groups.
