On Providing Survivable QoS Services in theNext Generation InternetAnotai Srikitja and David TipperDept. ofInformation Science and TelecommunicationsUniversity ofPittsburghPittsburgh, PA 15260 USADeep MedhiDept. ofComputer NetworkingUniversity ofMissouri - Kansas CityKansas City, MO 64110 USAAbstract— In this paper we present a comparative study of twoschemes to provide survivability for guaranteed QoS connections in apossible Next Generation Internet network architecture.scheme a QoS connection is provided standby backup resources on a dis-joint path by reserving resources on both the working and backup path.In order to reduce the amount of backup resources required a methodfor sharing backup resources when the working connections have dis-joint routes has been included. In the second scheme a dynamic searchfor restoration resources is conducted over a preplanned set of alternatepaths upon notification of a failure. A simulation based performancestudy shows that the first scheme results in much higher connectionblocking under normal operations, slightly faster restoration times, andlonger transient congestion times after fault recovery due to non-optimalbackup routing.In the firstI. INTRODUCTIONThe Next Generation Internet (NGI) will provide QoS-based services in addition to traditional best effort service. Acertain set ofusers (e.g., military) can be expected to demanda cost effective level offault tolerance forQoS-basedservices.Hence there is a need for techniques to ensure the survivabil-ity of certain services in NGI architectures. Note that someQoS services will likely follow fixed routes (e.g., GuaranteedService class in IntServ model [17]) and thereby be subject tosingle point failures (e.g., link failure) as in circuit switchednetworks.The subject ofproviding survivability in the face offailureshas been extensively studied for circuit switched networksand recently for ATM networks [3], [9], [18]. This includeswork on network design and capacity allocation, as well aswork on traffic restoration/network management algorithmsfor fault recovery. A variety of survivability techniques havebeen proposed for circuit switched and ATM networks at thephysical layer, logical layer and the traffic layer. At the phys-ical layer most of the work focuses on SONET rings or auto-matic protection switching both of which involve provision-ing idle spare capacity. At the logical and traffic layer severalapproaches for provisioning spare network capacity in meshtype topologies have been proposed along with traffic restora-tion techniques. It is well known that a mesh type topologytogether with traffic restoration techniques is more capacityefficient than physical layer approaches. The traffic restora-tion techniques typically differ in specification ofthe locationof rerouting, the rerouting algorithm and the reservation orsupported in part by NSF grant NCR 9506652 and DARPA under agree-ment No. F30602-97-1-0257nonreservation of resources in the event of a failure. Fromthe literature it is clear that no one approach is cost effectiveor optimal for all networks and a multi-layer survivability ap-proach with a combination of techniques is suggested. Suchsurvivability issues have received little attention in the contextofNGI architectures as the focus thus far has been to developa QoS based architecture.In this paper we investigate two schemes for survivabilityapplicable at both the traffic or logical layers in any packetswitched network that supports explicit path establishment.In the first scheme a guaranteed QoS service is provided re-served standby backup resources on a path which is disjointwith the working path.In order to reduce the amount ofbackup resources required, the method for sharing backup re-sources proposed in [4] is adopted. In the second approach adynamic search for restoration resources is conducted over apreplannedset ofalternate paths upon notification ofa failure.A comparative simulation based study of the performance ofthe two survivability schemes is presented. The study con-sidered both steady state and transient network behavior andshows that the first scheme results in much higher connectionblocking under normal operations, slightly faster restorationtimes and longer transient congestion times.In the next section, we discuss NGI architectures and thetwo survivability schemes studied in detail. In Section III,we present the results of our simulation based performanceevaluation. Lastly in Section IV we summarize our findings.II. SURVIVABILITY SCHEMES FOR NGICurently two different QoS-based services frameworks arebeing considered for the NGI namely: (1) the Integrated Ser-vice (IntServ) model and (2) the Differentiated Service (Diff-Serv) model. In the IntServ model, per connection QoS-basedservices are classified into three types: Guaranteed Service,Controlled-Load Service and Best-effort Service [17]. TheGuaranteedService class is intended to provide a guarantee ofbandwidth, a end-to-end delay bound and no packet loss, toapplications with a stringent real-time delivery requirement.The Controlled-Load service is aimed to be used for thoseclasses of applications that can tolerate some amount of lossor delay, i.e. adaptive real-time applications. Controlled loadservice provides a loose guarantee of service on delay andpacket loss. Best-effort Service corresponds to current Inter-net service in that no guarantees are made.
