Internet Routing Dynamics and NSIS Related Considerations

draft-shen-nsis-routing-00.txt

Charles Shen, Henning Schulzrinne, Sung-Hyuck Lee

IETF#61 – Washington DCNovember 2004

Outline An Internet Routing Dynamics Measurement

Measurement Methodology Summary of Results from NSIS Perspective

NSIS-Concerned Route Changes Typical NSIS Deployment Models Evaluation of Packet TTL Monitoring Route

Change Detection Conclusion and Next Step

Measurement Methodology

Traceroute end-to-end path characterization 24 Public servers located in US, Iceland,

Netherlands, Australia, Germany, Switzerland, Bulgaria, Sweden and Thailand.

Independent, exponential sampling interval 15/30min per-site (2.75h/11.5h per-path)

Selected paths with 10min fixed interval Between April and August 2004

Summary of Results (I) Route Prevalence and Route Persistence

Paths strongly dominated by a single route Significant site to site variation exists. Adaptive approach for NSIS and routing

Different Types of Route Changes Wide range of time / location scales Majority with no change of total hop count Route splitting and load balancing

Summary of Results (II) Accuracy of Measuring Path Characteristics

10-min fixed / 2-hour exponential interval Both capture the same number of routes, AS-paths changes The latter missed about half the number of routes, AS-paths

and changes. Essentially site-to-site variation – adaptive mechanism eg.

Refresh? Impact of multi-homing

AS level route changes and asymmetric routing An example:

change of main outgoing ISP from month to month but still occasionally use the previous ISP (10-30min) No change of incoming ISP

NSIS-Concerned Route Changes

Generic route change Inter-AS Intra-AS: Ingress-point, Egress-point, Mid-point

Deal with all generic route changes only in a full NSIS model

But a mixed NSIS deployment model more likely

NSIS-concerned route changes (NCRCs) Involving change of NSIS entities in the path Subsets of generic route changes

Typical NSIS Deployment Models

AS model: a central NE in each AS NCRCs equivalent to inter-AS route changes (RCs)

Entry model: ingress routers of ASes are NEs NCRCs - inter-AS and intra-AS ingress route changes

Border model: both ingress and egress routers of ASes are NEs NCRCs – inter-AS, intra-AS ingress and egress route

changes Edge model:

access routers of src/dst sites are NEs. NCRCs – inter-AS RC involving the first or last AS, intra-AS

ingress RC in the first AS, intra-AS egress RC in the last-AS.

TTL Monitoring Evaluation (I)

Description DS I DS II DS III

TTL-visible RCs 38% 25% 23%

AS level RCs 18% 8% 8%

TTL-visible AS level RCs

77% 83% 88%

Overall TTL-visible RCs not so promising Most concerned are non-trivial RCs Pretty good for AS level changes

TTL Monitoring Evaluation (II)

TTL method more effective in these models Ratio higher in sparser models Generic mixed model falls between these results

NSIS Model DS I DS II DS III

AS Model 77% 83% 88%

Entry Model 51% 41% 40%

Border Model 45% 39% 38%

Edge Model 74% 90% 92%

Conclusions and Future Work

A recent end-to-end routing measurement Different Route Changes require different

handling routing monitoring inside the network for frequent

yet local route changes caused by route splitting or load balancing.

simple packet TTL monitoring reasonably good for NSIS-concerned route changes in typical NSIS deployment models.

More route change detection methods to be evaluated.