Information Technology Reference
In-Depth Information
Memory use on a PE is determined by the following factors:
•
Number of VRFs
—Each VRF structure consumes a certain amount of memory.
•
Number of local VPN routes
—Memory use increases with the number of local VPN
routes.
•
Number of remote VPN routes
—Memory use increases with the number of remote
VPN routes.
•
RD allocation schemes
—These schemes affect how VPN routes are stored. You'll
read more about this topic in the section “Design Guidelines for RDs.”
•
Number of CE neighbors and type of connectivity
—Neighbor structures use
memory.
•
CE-PE protocol
—Different protocols use different structures and consume memory
differently.
•
Number of iBGP peers
—Memory use increases with the number of peers.
•
Number of global routing table entries
—Memory use increases with the size of
the table.
•
Hardware platform
—Hardware-specific structures use memory differently.
•
IOS release
—Different releases might store and cache the information differently.
If large numbers of routes, such as full Internet routes, are delivered in a VPN, the CSC
model should be used. As another example, separation of IPv4 and VPNv4 routes and the
use of RRs (discussed in the next section) reduce resource consumption. During capacity
planning for PE devices, keep the following points in mind:
•
The number of VRFs per PE is limited primarily by CPU, whereas the number of VRF
routes is constrained by available memory.
•
Do a baseline assessment before adding any VPN routes. Within the baseline, take into
account the memory uses of the IOS image, backbone IGP routes, Internet routes (if
any), and forwarding structures such as FIB and LFIB.
•
Assess the extra requirements added by VRFs, considering overhead memory per
VRF (about 60 to 70 KB) and memory use per VRF route (about 800 to 900 KB).
•
Leave an additional amount of memory (about 20 MB) as transient memory.
Route Reflector Designs with MPLS VPN
Route reflection can be effectively used to reduce CPU and memory use in an MPLS VPN.
An RR can selectively reflect routes between a group of PE devices with proper use of
filtering. In an inter-AS environment, for example, the use of eBGP multihop peering
between VPNv4 RRs reduces memory use in ASBRs.
