270 likes | 395 Views
Link Failures in IP Networks:. A Closer Look. Yashar Ganjali Sprint Advanced Technology Lab. August 5, 2003. Previous Work. Link Failures. Maintenance. Unplanned. 70 % of Unplanned. Shared: Router Related, Optical Related, …. Individual Link Failures. Our Work.
E N D
Link Failures in IP Networks: A Closer Look Yashar Ganjali Sprint Advanced Technology Lab. August 5, 2003
Previous Work Link Failures Maintenance Unplanned 70 % of Unplanned Shared: Router Related, Optical Related, … Individual Link Failures Link Failures in IP Networks: A Closer Look
Our Work • Goal: gaining a deeper understanding of the causes & characteristics of link failures • How: • Using SONET alarm logs • Focusing on high failure links/periods Link Failures in IP Networks: A Closer Look
IS-IS Failures, Feb. 2003 Link Failures in IP Networks: A Closer Look
Number of Failures per Day Link Failures in IP Networks: A Closer Look
Zoom In Link Failures in IP Networks: A Closer Look
Zoom In (Cont’d) Link Failures in IP Networks: A Closer Look
High Failure Periods • Periods with a large number of failures (HNP) • Periods in which a large percentage of links are down (HDP) Link Failures in IP Networks: A Closer Look
Number vs. Duration of Failures Link Failures in IP Networks: A Closer Look
High Failure Periods Link Failures in IP Networks: A Closer Look
Periods with High Number of Failures Link Failures in IP Networks: A Closer Look
High Failure Periods (Cont’d) • We can also classify high failure periods based on spatial distribution of the links failing in those periods. Link Failures in IP Networks: A Closer Look
Spatial DistributionJune 17, 2003 Link Failures in IP Networks: A Closer Look
Spatial DistributionFeb. 26, 2003 Link Failures in IP Networks: A Closer Look
Links with High Duration of Failures Link Failures in IP Networks: A Closer Look
Links with High Number of Failures Link Failures in IP Networks: A Closer Look
Matching IS-IS Failures with SONET alarms (SLOS) Link Failures in IP Networks: A Closer Look
Preliminary Results • About 58% of all failures match with a SLOS alarm. • High Failure Links have a higher correlation with SLOS alarms (85%) • Router Related Failures show much less correlation than other classes (0-29%). • Periods with high number of failures show more correlation than periods with high duration of failures. Link Failures in IP Networks: A Closer Look
Unmatched SLOS alarms Link Failures in IP Networks: A Closer Look
Matching SLOS alarms • Problem: A large percentage (45%) of SLOS alarms do not correspond to any IS-IS failures • Solution: Remove links which are not up or links which go down for ever • Result: Less than 2% of SLOS alarms do not correspond to any IS-IS failures Link Failures in IP Networks: A Closer Look
Matching IS-IS Failures & SONET alarms Link Failures in IP Networks: A Closer Look
Unmatched SLOS alarms(considering removed links) Link Failures in IP Networks: A Closer Look
Unmatched SLOS alarms • Spread over time • Almost half of them have a matching alarm (SLOS <-> SLOS cleared) • Minimum time between SLOS and SLOS cleared is 9 seconds. • More investigation??? Link Failures in IP Networks: A Closer Look
Research Direction • High Failure Periods: Study cascading effects, spatial distribution of failures • High Failure Links: Predicting future failures based on passed ones. • Network Availability: How do failures affect network availability? • ??? Link Failures in IP Networks: A Closer Look
You always pass failures on the way to success! Thank you! Link Failures in IP Networks: A Closer Look
Spatial-Temporal Correlation of Failures Link Failures in IP Networks: A Closer Look
Spatial-Temporal Correlation of Failures Link Failures in IP Networks: A Closer Look