Information Technology Reference
In-Depth Information
The output also contains the memory use for path attributes, community, caches, and so on,
depending on the BGP configuration and prefixes received from peers. Note that these
numbers are smaller than what are estimated (as shown in Table 2-6). This is because the
memory numbers in the
show ip bgp summary
output does not include the memory
overhead. The results in this case study were obtained directly from the output of
show
memory
, which includes all memory usage.
If BGP inbound soft reconfiguration is enabled locally, all denied routes are still retained
as receive-only routes, leading to higher memory use for the BGP RIB. Because the receive-
only routes are excluded from the best-path selection, they do not affect the memory use for
IP RIB and IP CEF. With the route refresh feature available since the 12.0 release, inbound
policy changes are updated dynamically to peers, so inbound soft reconfiguration is no
longer required. The route refresh feature is on automatically for supported releases. To ver-
ify whether it is supported, execute the
show ip bgp neighbor
command.
Memory use for caching route maps and filter lists is not considered in this test. For a
typical Internet router with 100,000 routes and six different BGP paths, this portion of
memory use could be in the vicinity of 2 MB, and the total BGP memory use is about
80 MB. Combined memory use for BGP Scanner, BGP I/O, and BGP Router process
maintenance is generally well under 50 KB.
In this case study, only static memory use is estimated for BGP. Static memory here refers
to the memory use when BGP is in a steady state—that is, when prefixes are converged.
However, BGP might use additional memory during convergence. This type of memory is
called
transient memory use
. The size of transient memory is difficult to track and can vary
according to factors such as how updates are sent and received, the state the BGP Router
process is in, and the IOS releases. For example, peer groups allow updates to be replicated
from a peer group leader to other members of the group, so less memory is needed to hold
the messages. Update packing is another method to reduce the number of update packets
sent to peers. These and other performance tuning techniques are discussed in detail in
Chapter 3.
The BGP Router process can be in one of three states, depending on the IOS releases, BGP,
and router states, with increasing functionality and memory use:
•
Read-only
—BGP accepts updates only from peers. It does not calculate the best path,
nor does it install routes into the routing table. This reduces transient memory use.
During initial router bootup, BGP is typically in this mode.
•
Calculating the best path
—BGP accepts updates and runs through the path-
selection process, which generally is associated with the caching of some structures
and thus increased transient memory use. This typically is the transition mode.
•
Read and write
—BGP accepts updates, calculates the best path, installs the routes
into the IP routing table, and generates updates to be sent to peers. More transient
memory is needed. This is the normal BGP mode.
