Information Technology Reference
In-Depth Information
system degradation. Thus, cache-based switching has scalability problems for Inter-
net core routers. As another example, the efficiency of NetFlow switching depends on
the flow's length. If there are large numbers of short flows, new entries are created
constantly, resulting in lower efficiency and performance.
•
It is possible for caches to grow larger than routing tables, such as when multiple
equal-cost paths exist. As a result, fast cache can consume significant amounts of
memory.
•
Periodic aging of the cache entries can consume large amounts of CPU time if the
cache is large.
•
Cache invalidation because of a route flap relies on process switching to repopulate
the cache with valid entries. When the route table changes, the affected old entries
must be invalidated, and traffic previously using the cache entries is process-switched
as the cache is rebuilt. If there are a large number of flaps, which occur frequently on
the Internet, significant cache invalidation is seen, reducing the effectiveness of the
cache-based switching mechanism. This can also result in contention for system
buffers and loss of control traffic, contributing to network instability.
•
Cache-based switching is unable to do per-packet load sharing at an interrupt level.
Because cache-based switching is entirely destination-based, load sharing occurs only
on a per-destination basis.
Cisco Express Forwarding
As discussed in the preceding section, although cache-based switching mechanisms
improve forwarding performance over process switching, their performance is nondeter-
ministic. Both process switching and cache-based switching are data-driven or demand-
driven. In other words, the switching components are in place only after the packets enter
the router, and they are removed when such packets are not being forwarded by the router.
If there are large numbers of packets with unpredictable patterns, switching performance is
degraded significantly. Obviously these switching paths are not scalable at the Internet level.
CEF was created to avoid the problems inherent in cache-based switching mechanisms. It
is designed to best accommodate the changing network dynamics and traffic characteristics
resulting from increasing numbers of short duration flows typically associated with
web-based applications and interactive TCP sessions.
CEF offers the following benefits:
•
Scalability
—CEF is topology-driven and relates closely to the routing table. CEF
also offers full switching capacity at each line card when Distributed CEF mode is
active. CEF supports hardware-assisted forwarding, necessary to offer line rate
switching on high-capacity line cards.
