International Journal of scientific research and management (IJSRM)

||Volume||4||Issue||08||Pages||4530-4535||2016||
Website: www.ijsrm.in ISSN (e): 2321-3418


Effect of Route Reflection on IBGP Convergence and an approach to reduce
convergence time
Santhosh S, M Dakshayini,

M.Tech, CNE Student

Dept. of ISE, BMSCE
[email protected]
Professor, Dept. of ISE,
BMSCE, Bengaluru
[email protected]

Abstract— Request for Comments (RFCs) for Border gateway Protocol (BGP) suggest that the network topology using BGP must
have full mesh of IBGP sessions to avoid routing loops. Routing loop is a condition where packets are routed between two or more
routers resulting in slow convergence and routing instability. Route reflection is one of the strategies to avoid full mesh of internal
BGP sessions between BGP speaking routers in an autonomous system. This paper focuses on demonstrating an effect of using
route reflection on convergence time in a small network. A mixed network scenario of external and internal BGP sessions is
considered for demonstration. Results of the simulation have shown that, route reflection can significantly increase the
convergence time avoiding full mesh. Also an approach to reduce the convergence time has been proposed. Proposed algorithm is
based on the principle of creating logical group of router reflectors. The same concept can be scaled for larger networks for
different scenarios.
packets cross multiple autonomous systems. This makes an
Keywords: IBGP full mesh, convergence time, route reflector, obvious choice of using autonomous system number as a
Convergence time reduction algorithm parameter to check for loop. A loop is detected when a router
finds multiple entries for the same AS number in the AS path.
I. INTRODUCTION AS path is the list of all AS numbers traversed by the packet.
Whereas IBGP cannot use this technique because routing
An autonomous system (AS) can be thought of as an entity updates does not go out of the AS. Only approach given by the
comprising of a number of network devices under a single initial BGP RFCs is to have full mesh of IBGP session amidst
technical administration. The network devices concerned with all the BGP speaking routers in the AS. This implies that a
are typically routers and layer three switches. For example network containing N routers need to establish and manage
consider a university campus having ten routers belonging to N*(N-1)/2 IBGP sessions, which can be very hectic and
various departments. All these routers are managed by the tumultuous for the network administrators as they need to do
university’s IT department. Here an autonomous system manual configuration on all the routers.
consists of ten routers. In most of the cases one of the routers is Rest of the paper is organized as follows. Section II discusses
connected to an ISP to provide internet service to the users or study model for route reflection. Section III focuses on the
Virtual private network with other universities. Alternatively impact on convergence time. Section IV illustrates an approach
two or more routers can be used to connect to the outside world to reduce convergence time when route reflection is used.
through multiple ISPs to provide resiliency and improve fault
tolerance.
BGP is conceded as routing protocol of the internet backbone. II. STUDY MODEL OF ROUTE REFLECTION
There are two well-known forms of BGP called IBGP (internal .
BGP – within an autonomous system) and EBGP (External A model considered for this work is as shown in figure 1. In
BGP -between multiple autonomous systems). BGP session this model, a selected router is made as router reflector (RR) in
between the two routers within a university campus is an an AS. All IBGP speaking routers in the AS form IBGP
example of IBGP. BGP session between a router in the sessions with route reflector only. All the non RR IBGP
university campus and ISP’s router is an example of EBGP. speaking routers are called route reflector clients (RRC). An
Each autonomous system is assigned and uniquely identified by RR receives BGP updates from an RRC and forwards it to
a number called AS number. other RRC or to an EBGP neighbor depending on the
It is important to prevent routing loops in any of the routing destination address.
protocol and BGP is not an exception. It follows different loop Route reflection gained a lot of popularity from the outset of
prevention strategies in EBGP and IBGP. In case of EBGP, the concept of preventing full mesh in IBGP. Vendors and

Santhosh S, Volume 4 Issue 08 August 2016 [www.ijsrm.in] Page 4530

 DOI: 10.18535/ijsrm/v4i8.05

operators started adopting this technology in their designs router ID of the source router. Router ID is in general
without detailed testing or analysis because of the reduced configured as highest loopback IP address on the router.
number of BGP sessions it results in. Cluster list contains a list of routers IDs. A router updates the
In figure 1 Router – R1 is configured as an RR. Remaining cluster list with its router ID if it is not found in the cluster list.
three routers are configured as RRCs. The packet is discarded otherwise.
Router reflection has some advantages – it reduces the
number of IBGP sessions drastically to N-1 where N is the
number of BGP speaking routers in the AS. It may also reduce
the operational expenditure to some extent i.e. addition and
deletion of a BGP session becomes easier for the administrator.
The size of the routing table and number of updates are also
reduced compared to full mesh.
These advantages come at a cost. RR can affect the
convergence time of BGP and it can even reduce the path
diversity. Path diversity refers to the scenario where a router
can select best path to a destination when multiple choices are
Figure1. R1 Configured as an RR available. A better path to a destination may not even be
considered just because it is not learnt by an RR. This can
affect QoS sensitive traffic. One of the simplest methods to
Multiple routers can be configured as RRs to improve fault avoid path diversity problem is to give meticulous attention to
tolerance at the RR level. If connectivity with primary route the placement of RR in the network design. A well placed RR
reflector RR1 and any of the RRCs goes down for some reason can strategically solve the path diversity problem.
such as link or interface failure, power failure etc. R2 takes the Effect on convergence time: using an example of a small
responsibility of RR and BGP communication is unaffected. network, effect of RR on convergence time can be studied. This
This is shown in figure 2. study is specific to the scenario considered and the results
cannot be generalized as there are various parameters which
can affect the convergence time. For the same scenario
convergence time is calculated using RR and full mesh for
comparison.

III. IMPLEMENTATION TO CALCULATE

CONVERGENCE TIME
Convergence time is one of the most important performance
metrics of a routing protocol. When there is a change in the
network topology, it takes some time for the update to reach all
the routers in the network. Alternatively convergence can be
thought of as the process of restoring a network back to its
normal state when the problems that caused the outage are
fixed.
In the considered example shown in figure 3 and figure 7
Figure2: Multiple RRs provide resiliency at the RR level. routers R1, R2, R3 and R4 are IBGP speakers with AS 299. R1
connects the AS to the external world.R1 has an EBGP session
Because of introduction of an RR in the AS, reachability with R5. Router R5 is in AS 399. R1 and R4 are chosen as the
information need to travel more than one hop to reach its farthest points in the network for implementation. Selecting
destination within AS. This could be a potential reason for a farthest points gives better approximation of convergence time.
loop if counter measures are not taken. Addition of two BGP
attributes namely ‘Cluster list’ and ‘Originator ID’ takes care of
the loop problem because of the extra hop. Originator ID is the

Santhosh S, Volume 4 Issue 08 August 2016 [www.ijsrm.in] Page 4531

 DOI: 10.18535/ijsrm/v4i8.05

Figure 3: Full mesh IBGP

connectivity to R5 through all possible paths. Now

convergence time can be measured as follows.
Figure 4 below shows the output taken at R4 proving PC2 is
reachable from R4 using BGP. A continuous ping from PC1 to PC2 is started when the
interface is turned up at R1 to bring back the connectivity to
life. Figure 5 shows the instant at which it is turned up. Ping
from PC1 to PC2 did not start to work immediately when
problem is fixed. After a lot of packer drops ping starts to
work. The round trip time for all ICMP packets are shown an
each line of the ping output. Sum of all round trip times given
an approximation of convergence time.

a. Continuous ping result at PC1 is shown in figure 6.

Complete output could not be shown as the logs are more than
a page. RTT for each packet is shown in the fourth field at each
Figure 4: routing table entry response. Calculated convergence time is approximately equal
to 20 seconds for full mesh topology
A network outage is simulated by deliberately shutting down
R1’s interface that connects to R5. This makes R4 loose

Figure 5: problem that caused the outage is fixed.

Santhosh S, Volume 4 Issue 08 August 2016 [www.ijsrm.in] Page 4532

 DOI: 10.18535/ijsrm/v4i8.05

Figure 6: continuous ping from PC2 to PC1

b. Same experiment is repeated for this network setup by IV. CONVERGENCE TIME REDUCTION PROTOCOL FOR
configuring R1 as route reflector. This is shown in figure 7. IBGP NETWORKS WITH RRS
For this scenario with R1 acting as route reflector, using the Considering a scenario where multiple RRs are used to provide
same technique described earlier convergence time is found to route reflection and these RRs connect to multiple ISPs, a
be approximately 30 seconds. method can be applied to reduce convergence time. BGP
This clearly shows that introduction of an RR caused 50% routers use keep-alive packets to inform its presence to its
increase in a very small network consisting of few routers. neighbors. Default value of keep alive timer is one minute.

Figure 7: IBGP with Router reflector

Santhosh S, Volume 4 Issue 08 August 2016 [www.ijsrm.in] Page 4533

A BGP speaker expects keep-alive packets from its active When a member of the group does not receive the keep
neighbors periodically every minute. When it does not receive alive from master for 20 seconds it is considered dead and
keep-alive from a particular neighbor it waits for 3 keep-alive control is switched over to the router with next highest
periods that is by default 3 minutes before terminating the BGP priority (back up RR). Backup RRs are made to wait only
session. If it is possible to alter this behavior, convergence can for one third of the standard BGP keep-alive period.
be made up to 3 minutes faster than usual. If back up RR also broadcasts its status to other routers, it is
BGP policies must be defined in ordered to make the failover possible to take care scenarios where both current master
to second RR when the connection goes down to the primary and backup router become unavailable at the same time.
RR. The amount of time for which it has to wait must be less
than default hold timer value. This calls for a very high degree The keep alive packets contain only the necessary
of robustness in the failover strategy among the RRs. information such as Master route reflectors RR_ID,
GROUP_ID, and priority value. This results in saving time
A logical group of all router reflectors can be created called when switching over from master to back up RR thus
RR_GROUP. An RR_GROUP is identified by using its unique resulting in reduced convergence time.
logical GROUP_ID. Each RRC does not connect to individual
RRs but it connects to RR_GROUP’s logical ID. Each RR in a V. CONCLUSION
group is assigned a priority based on the computing abilities of Exact value of convergence time depends on several
the router, its position in the network etc. The priority parameters such as size of the network, processing ability of the
assignment is done manually to avoid overhead of a master RR, strategic placement of the RR, number of RRCs,
election mechanism. throughput of the links connecting routers etc. When using RRs
in networks that require stringent QoS requirements it is very
This calls for necessity of a protocol that can effectively important to design the network to harness maximum
synchronize communication between route reflectors. The only performance. This paper provides a novel approach to reduce
goal of the protocol is to coordinate among all the route the convergence time by taking advantage of the BGP timers
reflectors in RR_GROUP and communication from RR clients and creating a logical group of route reflectors.
to RR Group. Each RR has an ID which is locally significant
within the RR_GROUP called RR_ID. The status of current ACKNOEDGEMENT
master RR is propagated using a small packet to all the backup The authors would like to acknowledge and thank Technical
RRs. This packet has very low overhead and includes only the Education Quality Improvement Program me [TEQIP] Phase 2,
status of the master RR and its RR_ID. Because of the small BMS College of Engineering and SPFU [State Project
size of the packet it can be sent more frequently not causing Facilitation Unit], Karnataka for supporting the research work
any effect on the network performance. When the master RR
status goes down, backup RR must pick up the role of master REFERENCES
RR in next update period. [1] Yuri Breitbart ,Minos Garofalakis , Anupam
Gupta , Amit Kumar ,“On Configuring BGP route
Reducing the hold timer to a suitable value and using the
reflectors” in conf rec Communication Systems
protocol explained above, a sufficient reduction in convergence
time can be obtained. Software and Middleware, 2007. COMSWARE 2007.
2nd International Conference , pp. 1-12
Proposed convergence time reduction Algorithm [2] Jong Han Park ,Ricardo Oliveira ,Shane
Amante , Danny Mcpherson ,“BGP Route reflection
Step 1 : Initialization phase: revisited” IEEE Communications Magazine , July
Each RR is given a RR_ID, it can be router’s loopback IP 2012 , pp 70-75
for instance. [3] Nick Feamster, Jennifer Rexford , “Network-Wide
An RR_GROUP is given an ID called GROUP_ID. The IP Prediction of BGP Routes” IEEE/ACM Transactions
addresses of the interfaces on all RRs that connect to the on Networking , 2007 , pp. 253-266
IBGP network are in the same subnet (IBGP LAN
[4] R. Musunuri,J. A. Cobb " A complete solution for
subnet).Then GROUP_ID can be an IP address from the
same IBGP LAN subnet provided the ip address is used on iBGP stability" Communications, 2004 IEEE
any of the physical interfaces. International Conference June 2004, pp. 1177 –
RR clients are aware of only the GROUP_ID. RR clients 1181
use GROUP_ID in a way similar to default gateway to [5] U. Bornhauser , “Root causes for iBGP routing
communicate with RRs. anomalies" Local Computer Networks (LCN), 2010
IEEE 35th Conference , pp 480 – 487
Step 2: Maintenance and failover phase: [6] H. Gobjuka , "Forwarding-loop-free configuration
The goal here is to ensure high availability. for IBGP networks" Networks, 2003. ICON2003. The
Master RR is selected based on the computing abilities of
11th IEEE International Conference , pp 31 – 37
the RR and its placement in the network.
Each RR is given a priority in the group. Active router at
[7] Li Xiao , Communications, 2003. ICC '03. IEEE
any point of time with the highest priority value becomes International Conference, pp 1765 - 1769 vol3
the master router in the group. [8] V. Van den Schrieck , "BGP Add-Paths: The
Low overhead keep alive packets are broadcasted by the Scaling/Performance Tradeoffs" , IEEE Journal on
master within the group periodically in short intervals.

Santhosh S, Volume 4 Issue 08 August 2016 [www.ijsrm.in] Page 4534

 Selected Areas in Communications (Volume:28 ,
Issue: 8 ), pp 1299 – 1307
[9] B. Wang , "The research of BGP convergence time" ,
Information Technology and Artificial Intelligence
Conference (ITAIC), 2011 6th IEEE Joint
International , pp 354 - 357

Santhosh S, Volume 4 Issue 08 August 2016 [www.ijsrm.in] Page 4535