Professional Documents
Culture Documents
Summarization
As the network size increases, the number of individual networks listed in the IP route table also
increases, as does packet size.
Routers cannot effectively handle many subnetworks, which leads to slowdowns, packet
losses and even crashes. That's why it's important to reduce the number of entries in the route
table, which is what route summarization accomplishes.
With route summarization, many routes are advertised with just one line in an update packet,
which not only reduces the packet size, but also allows more bandwidth for data transfer.
If a router needs to advertise 50 routes, it will need 50 specific lines in its update packet.
As these routes increase, the number of lines required also increases, expanding packet
size and the amount of bandwidth used. That means there will be less bandwidth
available for actual data transfer.
Route summarization enables multiple routes to be advertised with only one line in an
update packet, reducing the packet size and leaving more bandwidth for data transfer.
Also, each time a new data flow enters a router, it must identify which interface the
traffic must be sent out to. For this, it must perform a lookup in its routing table.
This process takes longer for large routing tables and requires more router central
processing unit (CPU) cycles to route traffic.
Route summarization can eliminate this problem by minimizing both the time required to
perform lookup and reducing the number of CPU cycles.
1. Reduces the number of entries in the route table, which reduces the load on the router
and network overhead for routing protocols;
2. Minimizes latency in a complex network, especially when many routers are involved;
3. Reduces or eliminates unnecessary routing updates after part of the network
undergoes a change in CPU cycles topology;
4. Hides instability in the system behind the summary that remains valid even in the
absence of summarized networks;
5. Saves memory since routing tables will be smaller in size;
6. Helps save bandwidth as there are fewer routes to advertise; and
7. Reduces processor workloads and saves, since there are fewer packets to process and
smaller routing tables to work on.
2. Forwarding traffic for unused networks. If the router doesn't find a matching destination in its
routing table, it will start dropping traffic, leading to data loss. Also, the summary route may
cover unused networks. The router that has a summary route will forward traffic to the
router that advertised the summary route.
Note:
To avoid suboptimal or incorrect routing and to prevent routers from inaccurately advertising networks or
duplicating other routers' advertisements, it's important to design networks with summarization in mind.
Advance planning and leaving room for future network growth can help with the design of a scalable
network that supports route summarization.
Let’s revisit the previous IPv6 deployment in Figure 11-20, but now summarize all the loopback addresses 2001:db8:0:1/128, 2001:db8:0:2/128,
and 2001:db8:0:3/128) along with the peering link between R1 and R2 2001:db8:0:12/64) on R2. The configuration would look as shown in
Example 11- 28.
Source : https://www.ciscolive.com/
Table 11-6 lists the bits needed for summarization, the IPv6 summary
address, and the component networks in the summary range.
Source : https://www.ciscolive.com/
The summarized network range must be changed to 2001:db8::/58 for summarization of the 2001:db9:0:23::/64 network to occur.
Example 11-30 shows the configuration change being made to R2. Example 11-31 verifies that the 2001:db8:0:23::/64 is now within the
aggregate address space and is no longer being advertised to R1.
Source : https://www.ciscolive.com/
Among many reasons for the wide adoption of SRv6 uSID technology is ultimate scalability. SRv6 uSID currently
provides full feature parity with SR-MPLS but in a much simpler manner, and with much higher scalability. The key
concept for infinite scalability is the applicability of classless routing (CIDR) to SRv6 uSID networks.
Let’s say we have a midsized network with 30k routers. It is obvious that we cannot handle such a network as a single
IGP domain.
We must split it into multiple IGP domains. Either using the hierarchical structure of IGP protocols (ISIS levels or OSPF
areas) or even using different IGP processes.
For simplicity, we split our network into 30 domains with 1000 nodes each. As we need to maintain any-to-any
connectivity, an obvious option is to redistribute all SRv6 locators everywhere. But IGP protocols have their scalability
limits as well. Attempt to redistribute all locators across would reach IGP limits.
SRv6 offers a very elegant solution to that problem: summarization. Every border router will propagate a few
summary prefixes instead of all locators.
Source : https://www.segment-routing.net/demos/upa/
Figure 1 - Summarization
In Figure 1 we can see an example of summarization. 1000 /48 Locator
prefixes are summarized into the core as four /40 networks. As a result,
in the core there will only be 120 summary routes instead of 30k, while
still providing any-to-any connectivity. 120 networks in a single domain
are simple to handle for any IGP routing protocol and easy to handle for
any HW platform.
i ia fccc:cc00:2011::/48
i ia fccc:cc00:2012::/48
Once we trigger the PE11 failure, IGP deletes the locator prefix of PE11 and PE1 triggers BGP
PIC.
This solution can only be used for small networks, where the IGP scale allows all locators to be flooded within the domain.
After configuring the summary prefix, PE1 no longer receives individual locators of domain 2:
Therefore, the IGP of domain 1 no longer receives PE11 failure notifications and BGP PIC can’t be triggered anymore on PE1.
In this measurement, the convergence was very slow (more than 50 seconds) due to the delay of BGP detecting and propagating the failure.
This is the sequence of events:
• The Route Reflector detects the failure of the BGP session to PE11
• The Route Reflector sends BGP withdraw messages to PE1 for all prefixes, one by one
• PE1 reprograms the FIB entry for each prefix
• The overall convergence time depends on the number of prefixes. The more prefixes the longer the convergence time will be. But we
need fast BGP PIC convergence even in very large networks where summarization is used!
The unreachability property of the prefix is carried by using an “unreachable” metric, which is already part of the ISIS
protocol definitions (According to RFC5308, any prefix advertised metric larger than MAX_V6_PATH_METRIC 0xfe000000
must not be considered during path computation and can be used for other purposes). Thus, UPA doesn’t require any
protocol extension.
The figure below shows the example network in stable state. The IGP of PE11 advertises its LSP with locator /48 into
domain 2. The ABR receives this LSP and advertises the /40 summary prefix into domain 1. PE1 receives this summary
prefix that provides reachability to PE11.
The goal of UPA is to notify about unreachability of prefixes so routers that are part of a remote domain can act upon this
notification. Thus, UPA prefixes are not intended to be persistent.
After some period, the ABRs automatically withdraw the UPA. This time is to allow full BGP convergence and is configurable.
For UPA to work, only the ABRs and ingress PEs routers need to be upgraded. All intermediate nodes will flood the UPA
prefix seamlessly.
Now we can simulate a PE11 failure and measure convergence using IXIA.
Figure 7 - Failure with summarization and UPA
We can see that the convergence time is exactly the same
as without summarization, precisely 353 milliseconds. This
is the sequence of events during that time:
1. The IGP in domain 2 detects PE11’s failure
2. The IGP floods the failure throughout domain 2
3. The ABR receives the failure information, generates a
UPA for PE11’s prefixes and sends it into domain 1
4. The IGP in domain 1 floods the UPA
5. PE1 receives the UPA and triggers BGP PIC – switching
to all preprogrammed backup paths