Stability Issues in OSPF Routing

flutteringevergreenΔίκτυα και Επικοινωνίες

29 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

57 εμφανίσεις

January 31, 2002

EECS
-

UC Berkeley

Stability Issues in
OSPF Routing

Basu and Riecke

George Porter

NETREAD, Spring 2002

January 31, 2002

EECS
-

UC Berkeley

Motivation/Overview


What is OSPF
-
TE? Stability?


Processor, route flaps, convergence


Subsecond HELLO timers


Link State Advertisement (LSA)
synchronization avoidance

January 31, 2002

EECS
-

UC Berkeley

OSPF
-
TE


Open Shortest Path
First, Traffic
Extensions


Link State
Advertisements (LSA)
sent periodically, and
when network changes


Flooding used to
distribute


RSVP: soft state


CR
-
LDP (constraint
based): hard state

January 31, 2002

EECS
-

UC Berkeley

Issues in stability


Convergence time


Processor load


Number of route flaps


Node failures


Message loss


Triggering thresholds


Subsecond Hello timers


Refresh Packets

January 31, 2002

EECS
-

UC Berkeley

Multiple node failures


Processor


1 failure: no effect


5 failures: 50%


10 failures: 100%


Route flaps:


OSPF affected by
as few as 5 failures


Convergence times:


OSPF robust in this
regard


TE overhead about
20%

January 31, 2002

EECS
-

UC Berkeley

Message Loss


Processor:


2% and 10% show
little effect (at
10%, OSPF thinks it
has fewer
neighbors, and so
traffic goes down)


5% utilization is
affected


Route flaps:


Little effect at 2%
and 5%


At 10%, there is a
huge amount of
flapping

January 31, 2002

EECS
-

UC Berkeley

Convergence time


Hello messages are usually sent every 10
seconds


Propagation time given Hello timer of


10 sec


33 sec


500ms


3 sec


250ms


1 sec


Processor load for subsecond Hellos is
basically the same (as long as over 275 ms)

January 31, 2002

EECS
-

UC Berkeley

Refresh Synchronization


Based on the LSA timers, periodic
bursts of traffic can develop


Proposal:


Introduce randomization


Zinin mechanism (but why? They said it
didn’t have much of an effect…)

January 31, 2002

EECS
-

UC Berkeley

Summary


OSPF with TE is pretty stable


TE has little effect on convergence time


OSPF resilient to message loss


Setting HELLO timers to <1 sec is a good
idea (if over, say, 300ms)


Randomization can prevent bursts of LSA
traffic