Information Technology Reference
In-Depth Information
A best-practice guideline during capacity planning is to increase another 20% of the static
memory use to account for the transient memory and any other variables. Another number
to watch out for is the minimum largest block of DRAM available in the system. If that
number is 20 MB or less, more resources are needed.
All best network/path entries in the BGP RIB are installed into IP RIB, resulting in memory
use for NDB and RDB structures. If a major network is subnetted, in fixed or variable
length, an additional entry is created in IP RIB for the major network. Each entry uses 1172
bytes of memory, depending on the IOS release. The memory use for subnetted entries is
shown as internal in
show ip route summary
. This value is the total number of entries in
show ip route
, with
prefix
subnetted or variably subnetted. Because only two to six major
networks were used in this test, the memory use for subnetted entries in IP RIB is less
than 7 KB.
Another important contributor to BGP memory use in IP RIB and IP CEF that is not con-
sidered in this test is BGP load sharing. By default, BGP installs one best path into the IP
RIB. With BGP multipath, multiple entries per BGP prefix may be installed into the IP rout-
ing table, resulting in increased memory use by IP RIB and IP CEF.
BGP prefixes installed in the IP RIB are populated in the FIB table. Memory allocation for
IP CEF is generally in line with what is reported by
show ip cef summary
. For line cards
running dCEF, this is the only memory use for BGP, because line cards do not maintain the
BGP RIB or the IP RIB. Besides the number of prefixes, memory use by CEF is also related
to prefix length. For example, if the prefix is /16, the memory use for this prefix is 1 KB on
top of the 1 KB used by the root of mtrie. If the prefix is /24, another 1 KB is used. If the
prefix is longer than /24, another 1 KB is used. Internet prefix distribution generally shows
9% for prefixes /16 and shorter, 83% for prefixes between /17 and /24, and 8% for prefixes
longer than /24. With the goal of establishing a simple method without losing accuracy, the
prefixes used in this test were all /24.
Summary
This chapter started by examining the relations between the control plane and the forward-
ing plane; both are fundamental functions of a router. As a routing protocol, BGP is a part
of the control plane. However, BGP's performance can be affected by the forwarding
plane's performance, because both planes might compete for the same resources, such as
CPU and memory. BGP processes in Cisco IOS software were discussed, with emphasis on
memory use and the interactions among the processes. The case study provided a simple
method to estimate BGP memory use in Cisco Internet routers. To build a solid foundation
for the rest of the topic, this chapter reviewed some of BGP's essential components. These
include BGP attributes, path selection, capabilities exchange, iBGP, BGP-IGP routing
exchange, and RIB.
