Cisco Express Forwarding Help
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
03-22-2013
03:05 AM
- last edited on
03-07-2019
12:25 PM
by
NikolaIvanov
Hi all,
I'm trying to understand a little about CEF, more specifically how it ties in with the use of CFC or DFC cards.
Would anybody be able to provide me a brief & simple summary of why you wouldn't by default use CEF on any network (I've seen it described as something used on large core networks and the internet)?
I'm also keen to understand it's relationship with distributed forwarding cards, whether it's compulsory to use with them or quite how it hangs together.
Any advise would be much appreciated.
Thanks,
Neil.
- Labels:
-
Other Switching
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
03-22-2013 03:26 AM
Hello Neil,
in a C6500 multilayer switch CEF is a necessary and core part: all the multilayer switching is based on CEF.
There is no alternative way to use it.
The supervisor builds the master copy of the CEF table that is hosted on the PFC. All DFCs download a copy of the CEF table for local use.
The DFCs make the linecard intelligent enough to take autonomous forwarding decisions having the whole CEF stored in memory.
The CFC provides just a way to interconnect to the switching fabric, but it does not have intelligence to take autonomous decisions on how to forward a packet. Actually the PFC is consultated and it provides info on how to forward the packet based on the master copy of CEF table.
On the other hand a linecard with DFC does not need to ask to the PFC how to forward a packet.
This provides better scalability and make the device much more powerful in terms of throughput and packet per second.
The drawback of DFCs is that they are hardware based, and the memory size leads to a maximum number of CEF entries that can be processed in hardware. Non XL DFC models support only 256000 entries that are not enough to deal with a full internet table that is in the order of 420,000 routes. XL models must be used and also on the supervisor model (it must be XL version too for the PFC) for building multilayer switches for dealing with full internet tables.
All these communications between linecards and supervisor happen out of band on dedicated internal links in the so called Ethernet out of Band Channel (EOBC),
Because CEF is topology based, even the first packet of a flow ( defined by its IP source address, its IP destination address) does not need to create an interrupt to the main cpu as it happens with fast switching or with older multlayer switching implementations that were traffic driven.
>> why you wouldn't by default use CEF
Generally CEF is used and recommended. On routers there are few cases where disabling CEF on specific interfaces is needed for some features.
Hope to help
Giuseppe
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
03-22-2013 03:32 AM
Giuseppe,
Thank you so much, very clear and exactly what I was looking for.
Very much appreciated.
Neil.
