25Apr/1117

Exercise – BGP and IGP interactions

A company  has a site  in Europe  and another one in Mexico. They are interconnected by  a private leased line. Each site is locally connected to an ISP (ISP-EU in Europe and ISP-MX in mexico). Border routers are respectively ER-EU in Europe and ER-MX in Mexico. Providers have allocated prefixes α in Europe and β in Mexico. Providers are doing in-gress filtering (i.e. a packet with a source address non allocated  by a provider is discarded by the first provider router).

In  the  rest of the  document, a  host A  (resp.  B)  will designate a host  in Internet (outside company network) close to the ISP-MX (resp. ISP-EU) with a metric linked to BGP announcement. C (resp. D) will represent an equipment in the Mexican (resp. european) part of the company network.

Inside the site, an IGP routing protocol allows all routers to know all internal prefixes.

AS Graph of the network description

 

Question 1 What are the differences between IPv4 and IPv6 in term of multi-homing.

Question 2 What is α and β nature? (PI or PA)?

Question 3 Is   it   possible   to   assign   private   IPv4   prefixes   to   the   router’s   interface managing the link between Europe and Mexico?

Question 4 Will prefixes α and β be stored as is in the core muting tables?

Question 5 Does the company need a public AS number to establish  a link with BGP providers routers? Is it necessary to establish a iBGP peering between both company BGP routers?

Question 6 How will react the Mexican ISP, if the Mexican site announces all the prefixes collected by its IGP?

Question 7 Do  the border routers have to announce inside the company network the totality of prefixes learnt from the IGP? What should be announced?

Question 8 In this configuration, why it is not necessary to send default routes from one site to another?

The company makes some agreement with ISPs to change in-gress filtering rules. They now include all the prefixes allocated by the different ISPs (note that this is not in the policy of current ISPs) .

Question 9 Is it useful to announce the default route over the atlantic link.  If the Mexican ISP is unreachable: can C join A? can A join C?

Question 10 What should  do   the  company  to  allow full connectivity, even  in  case  of failure of one of its ISP?

We   suppose   now   that   ISPs   agreed   to   announce   through   BGP,   prefixes allocated by other providers.

Question 11 What are the consequences on core network routing tables?

Question 12 Which   parameters  have   the   greatest  impact   on   route   selections:   prefix length or AS path length?

Question 13 Give a simple solution that force the traffic leaving to the Internet to go through the atlantic link only in case of failure of one ISP.

RFC 226O proposes one alternative :

When an  enterprise  border  router   connected  to  a particular   ISP determines   that the   connectivity between the  enterprise  and the Internet is up  through all of   its ISPs,   the router   advertises   (to  the border  router  of  that ISP)   reachability  to  only the  address  prefix that  the  ISP  allocated to   the   enterprise.   This way  in a  steady state routes  injected by the  enterprise  into  its  ISPs  are  aggregated by these   ISPs,   and  are  not propagated  into   the  “default-free”  zone  of the  Internet.

When an enterprise border  router connected to  a particular ISP determines   that the   connectivity between the  enterprise  and the Internet through one   or more  of  its   other  ISPs  is   down,   the   router starts  advertising reachability to   the  address prefixes   that was allocated by these ISPs to the   enterprise.   This would result in injecting additional   routing information into  the   “default-free”  zone of  the  Internet.   However,   one   could  observe  that the probability  of all multi-homed  enterprises  in  the   Internet concurrently losing connectivity to   the   Internet through one   or more  of  their  ISPs  is fairly small.     Thus   on average   the  number  of   additional   routes  in  the “default-free”   zone   of  the  Internet  due   to multi-homed  enterprises  is expected to be   a small  fraction of   the  total  number  of   such enterprises.

Question 14 What is  a “default-free” zone, Why  this solution reduces the number of entries in core routing tables ?

RFC states :

One   such possible  mechanism could be provided by BGP   [RFC1771].   In this  case border   routers within the   enterprise would have  an  IBGP  peering with each other. Whenever  one border   router  determines  that the  intersection between the  set of   reachable   destinations  it receives  via  its EBGP   (from  its directly connected ISP)   peerings and the   set of reachable destinations it receives  from another border  router   (in the   same enterprise)   via  IBGP   is  empty,   the  border  router would  start advertising to   its  external peer  reachability  to  the  address  prefix that was  allocated to  the   enterprise by  the  I5P  connected to   the other  border  router.   The   other  border  router  would  advertise   (via IBGP)   the   address prefix   that was  allocated to  the   enterprise by  the ISP  connected to  that router.   This   approach is known as   “auto  route injection”.

Question 15:  Is the traffic recovery instantaneous?

Question 16 Can  this method be used when the connectivity problem concern the ISP with the other ISP?

Question 17  Why this method cannot be used when the connectivity problem is located near by  the site (for example, the link between the  ISP and  the site  is down?)

Question 18 What method is currently used to allow multi-homing, why are its advantages compared to the solution we have studied.


This content is published under the Attribution-Noncommercial-No Derivative Works 3.0 Unported license.

Tags: , , , ,

17 Responses to “Exercise – BGP and IGP interactions”

  1. admin says:

    Question 1 What are the differences between IPv4 and IPv6 in term of multi-homing.


    When this exam was written, people believed in the fact that in IPv6 each providers will allocate a different prefix and the host would get two global addresses. But this lead to some difficulty in the choice of the source address. If routing send packets to the other provider, they will be dropped due to in-gress filtering. This is still a research problem, and currently multi-homing in IPv4 and IPv6 is solved the same way using Provider Independent addresses (PI).

    The rest of the exam will study if we can do multi-homing with PA (Provider aggregable) prefixes, but we will see that it is not so easy.
  2. admin says:

    Question 2 What is α and β nature? (PI or PA)?


    These prefixes are allocated by each provider, so they are Provider Aggregable prefixes.

  3. admin says:

    Question 3 Is   it   possible   to   assign   private   IPv4   prefixes   to   the   router’s   interface managing the link between Europe and Mexico?

    Yes without any problem, the only problem with private addresses allocation to router is management (and traceroute), here it will be done locally. 

  4. admin says:

    Question 4 Will prefixes α and β be stored as is in the core muting tables?

    No, since it is PA prefixes, they will be aggregate by the provider and a shorter prefix (we will note α– and β–) will be announced to the rest of the world.

  5. admin says:

    Question 5 Does the company need a public AS number to establish  a link with BGP providers routers? Is it necessary to establish a iBGP peering between both company BGP routers?


    No it is not needed. Public ASN are needed if they appears in the AS Path. Here, since the company is not doing transit traffic, it can use a private ASN that will be withdrawn by its ISPs.

    For the same reason, it is not necessary to do iBGP between routers, since they don’t have to re-announce something learned from the Internet

  6. admin says:

    Question 6 How will react the Mexican ISP, if the Mexican site announces all the prefixes collected by its IGP?

    The mexican ISP will reject prefixes it has not assigned (french prefix and private prefix)

  7. admin says:

    Question 7 Do  the border routers have to announce inside the company network the totality of prefixes learnt from the IGP? What should be announced?

    almost the same question as question 6. The border router should just announce the prefix assigned by the ISP it is directly connected to.

  8. admin says:

    Question 8 In this configuration, why it is not necessary to send default routes from one site to another?


    If one ISP fails, the traffic will be redirected to the other ISP, but traffic will be discarded since the source address will be different from the one expected.

  9. admin says:

    Question 9 Is it useful to announce the default route over the atlantic link.  If the Mexican ISP is unreachable: can C join A? can A join C?


    This is useless since this will also traffic to go through the other ISP to the destination, but the answer will be sent to the failing ISP and will never reach the source. C will join A but A will not be able to join C.

  10. admin says:

    Question 10 What should  do   the  company  to  allow full connectivity, even  in  case  of failure of one of its ISP?


    One solution is to use a private addressing pan internally and NAT the traffic with the exiting provider public address. This work well if there is no publicly accessible servers inside the company. The failure of one ISP will not be transparent since all connection will be broken

  11. admin says:

    Question 11 What are the consequences on core network routing tables?

    PI prefixes are stored without any aggregation in core routing table, so this lead to increase the table size.

  12. admin says:

    Question 12 Which   parameters  have   the   greatest  impact   on   route   selections:   prefix length or AS path length?


    Prefix length is taken in priority.

  13. admin says:

    Question 13 Give a simple solution that force the traffic leaving to the Internet to go through the atlantic link only in case of failure of one ISP.

    If we announce β prefix on the french ISP, the French ISP will announce a longer prefix than the mexican ISP (which is doing aggregation by announcing β–), so all the traffic will be directed to the french ISP. To avoid with both ISP should announce the same prefix length. 

    • guest says:

      dans le cas où on suppose que l’on utilise un préfixe PI (dans cette situation, il n’ y a plus de problème lié à la longueur du préfixe annoncé puisque les deux ISP vont annoncer le même préfixe) , si le AS de A reçoit deux annonces différentes pour le même préfixe, l’AS va garder la meilleure annonce, c’est celle qui correspond à un AS-PATH minimal, si le lien à travers l’ISP mexicain tombe en panne, A ne pourra plus joindre C? 

  14. admin says:

    Question 14 What is  a “default-free” zone, Why  this solution reduces the number of entries in core routing tables ?


    A “default-free” zone is a zone where there is no default prefix (0/0 or ::/0) in the router FIB. Routers know all the prefixes available on Internet.

  15. admin says:

    Question 15:  Is the traffic recovery instantaneous?


    No, it needs some time to propagate prefixes using BGP

  16. admin says:

    Question 16 Can  this method be used when the connectivity problem concerns the ISP with the other ISP?


    Yes, in that case the PA prefix will disappear from BGP routing table, the router will though BGP learned about other side prefix and will announce the prefix.

Leave a Reply

Your email address will not be published. Required fields are marked *