Post-Fault Restoration in Multi-Domain Networks with Multiple Failures



  Post-Fault Restoration in Multi-Domain.pdf (Size: 452.67 KB / Downloads: 0)

INTRODUCTION

Network survivability is a very well-studied area and a
host of solutions have been developed for IP-based multiprotocol
label switching (MPLS) and optical generalized
MPLS (GMPLS) networks. Most notably, these include preprovisioned
protection strategies for working/backup path
computation as well as post-fault restoration strategies; see
detailed survey in [1]. Furthermore, with expanding
deployments, a variety of multi-domain protection schemes
have been developed for larger national backbone networks
[2]-[6]. For example, SONET/SDH interconnection strategies
have been extended for “localized” dual/multi-homing link
protection between border nodes in optical dense wavelength
division multiplexing (DWDM) networks [3]. Alternatively,
“globalized” MPLS/GMPLS algorithms have also been
studied for diversified working/backup path routing using
sequential/parallel computation [5],[6].


ENHANCED CRANKBACK RESTORATION

A novel multi-domain crankback restoration solution is
now presented, extending upon the “working-mode”
crankback solution of [17]. The approach assumes realistic
settings with full intra-domain link-state routing and more
scalable path/distance-vector routing at the inter-domain level.
Each domain is assumed to have a PCE entity [19] with full
access to interior and exterior routing databases. This entity
plays a key role in crankback recovery as it resolves next-hop
domains. Meanwhile, all setup signaling is done using new
crankback extensions for RSVP-TE [15].


Multi-Domain Crankback: Restoration Mode

Now consider the case of crankback restoration under
multi-failure scenarios with combination of node and link
faults. Here it is assumed that a failed node will bring down
all of its constituent links, i.e., lead to multiple page link failures.
As a result, any working node connecting to a failed node will
quickly be able to resolve such occurrences via rapid lower
layer detection mechanisms (which are out of the scope here).
Hence with the failure notifications, appropriate crankback
restoration procedures can be initiated by the working nodes
bordering a failed region, detailed in the pseudo-code
description in Figure 2. This extends upon [17] by also
introducing RSVP-TE PATH_ERR message processing.


PERFORMANCE EVALUATION

The performance of the proposed multi-domain post-fault
restoration scheme is tested using specially-developed models
in OPNET ModelerTM, i.e., a modified NSFNET topology
(with nodes replaced by domains) with 16 domains/25
bidirectional inter-domain links, as shown in Figure 3. In this
topology, all intra and inter-domain page link rates are set to 10
Gbps and the individual domain sizes average 7-10 nodes and
correspond to metropolitan areas, i.e., cities. The NSFNET
topology is chosen as it is very representative of a nationwide
backbone and is widely used in many studies. Furthermore,
all requests are generated between random nodes in random
domains with mean holding times of 600 sec (exponential)
and variable inter-arrival times (as per desired load).


CONCLUSIONS
This paper proposes novel crankback solution for postfault
restoration in large multi-domain backbone networks
under multiple failures, i.e., WMD-type attacks. The scheme
uses an enhanced next-hop domain selection strategy and
incorporates full crankback history tracking to improve the
effectiveness of the recovery process. In addition, a dual
crankback counter approach is introduced to limit the number
of intra/inter-domain retry attempts and both end-to-end and
intermediate restoration modes are supported.