210 likes | 221 Views
This article discusses the routing policy lessons learned from TEIN2 and CERNET, including topics such as BGP experience, global transit, load balancing, peering, business model, performance, and symmetry.
E N D
Lessons LearnedTEIN2 and CERNET Xing Li 2007-01-22
Outline • Introduction • TEIN2 routing policy • CERNET BGP Experience • Lessons learned
Simple Case (where BGP can handle things easily) • Global transit • To tier 1 or tier 2 commodity networks • Care the aggregation • Care the load balancing • Don’t care the symmetry • Peering (no transit, except for the down streams) • To domestic ISPs (bi-literal or via IX) • Care the business model • To academic partners • Care the performance • Care the symmetry
Complicated Case (where BGP cannot handle things easily) • Global transit • To tier 1 or tier 2 commodity networks • Care the aggregation • Care the load balancing • Don’t care the symmetry • Academic transit • To multiple transit backbones within academic scope • Care the aggregation • Care the load balancing • Care the performance • Care the symmetry • Etc. • Peering (no transit, except for the down streams) • To domestic ISPs (bi-literal or via IX) • Care the business model • To academic partners • Care the performance • Care the symmetry
Two Steps to Implement the Policy • Identification • IP prefix • AS path regular expression • Community tag • Path selection • AS path (inbound and outbound) • Local-preference (outbound) • More specific (inbound)
The Principle of Routing Design for the TEIN2 network • To provide interconnection among TEIN2 partners and between TEIN2 partners and EU NRENs. • To provide back-up paths within the TEIN2 network and/or via partner networks for service resilience when possible. • To provide a flexible and transparent routing policy to TEIN2 NRENs. • To avoid being selected by GÉANT, Abilene and other R&E networks outside TEIN2 as the preferred transit network. • To minimize the adjustment of the external peers’ routing policy outside TEIN2 networks, e.g. GÉANT and APAN.
TEIN2 Routing Policy • Enable additive community tagging to mark the prefix announcements. • Adopt AS number prependingas the preferred BGP policy for TEIN2 traffic adjustment within TEIN2 backbone. • Use ingress AS number prepending for outbound traffic adjustment, including traffic from TEIN2 POP to NRENs, GÉANT and APAN. • Use egress AS number prepending for inbound traffic adjustment, including traffic from NRENs, GÉANT and APAN to TEIN2 POP. • May useLocal-Preference amendment as the last resort of mechanism for fine tuning on TEIN2 traffic over the backbone.
Internet 3 G 10G CNGI Peering Domestic Peering DRAGONTAP 12G CNGI-BJIX DRAGONLIGHT 155M 100M 1G 622M 155M 2x155M HARNET TEIN2 ASNET STARLIGHT APAN KOREN CERNET CERNET 2 CERNET Peering
CERNET Routing Policy • Outbound • Use AS number prepending if possible • Heavily useLocal-Preference • Enable additive community tagging to mark the prefixes • Inbound • Use AS number prepending if possible • Announce more specifics • Enable additive community tagging to mark the prefixes
Lessons Learned (1) • The nature of BGP is reachability • Stupid routing happen • Policy based routing makes thing very complicated • The routing and topology are very dynamic environment • The key words are: simple, open and controllability • For transit network • Simple • Open • For NRN • Simple • Controllability • Why did not include this policy before the earthquake? • Because it is a NP problem and there are many contradict requirements • Mission impossible • What should be the solution?
Lessons Learned (2) • It seems that we still need to do a lot manual BGP policy adjustment, case by case with the help of • Multi-site collaborations • Routeviews • We have to compare the routing table with the end-to-end performance matrix • dvping tool