Dynamic Allocation of Resources in VIOLA Abstract for VIOLA-Workshop Christoph Barz, Markus Pilz, Wolfgang Moll University of Bonn,, Römerstraße 164, 53117 Bonn, Germany {barz,pilz,moll}@cs.uni-bonn.de Keywords: Advance Reservation, Signaling, MPLS, GMPLS, Optical Testbed The German research project VIOLA (Vertically Integrated Optical Testbed for Large Applications) which is funded by the German Federal Ministry of Education and Research aims at the development of new mechanisms for dynamic and advance user bandwidth allocation and reservation in a heterogeneous multi vendor network infrastructure. In this context recent signaling mechanisms and network services will be evaluated as a basis for the X-WiN network, the next generation research network in Germany. Apart from the test of multi vendor network equipment and the development of software tools for bandwidth provisioning, the enhancement and test of Grid applications is another major focus of the project. This will allow for a realistic test scenario for the VIOLA network equipment. The Grid applications developed and tested in the VIOLA project have demanding requirements to the network. It is envisioned that requests for multiple Gbit/s need to be fulfilled in order to leverage these applications. Even in a high speed network like the VIOLA testbed resources are not overprovisioned. Instead they are coordinated by reserving them for a certain period. Connectivity requirements are one to one and one to many, often in an asymmetric manner. High demands for bandwidth may include long periods without network utilization. Other applications are more interested in short latencies and small jitter or in a combination of those. These requirements must be taken into account when providing a service access point (SAP) for Grid middleware and network users. In a heterogeneous network like the VIOLA testbed the integration of SDH, LAN/WAN Ethernet, MPLS and GMPLS nodes represents a challenge to future networks and upcoming migration scenarios, e.g. multiple technologies (TDM and packet switching), multiple layers and domains (MPLS and GMPLS). On this basis different layering models (Multi Layer and Multi Region Approaches) and network services like Layer 2 VPN, Layer 3 VPN and Diff- Serv are under examination to be offered to VIOLA customers. A network resource manager needs to reconcile the requirements of the Grid applications with the traffic engineering capabilities of the network devices. The resource manager has to introduce an advance reservation mechanism as network devices lack the concept of. This mechanism should be applicable for single domain reservations as well as for a network reservation involving multiple domains. Inter domain reservation can be used with several signaling options and path computation techniques. In addition, a network resource manager should support AAA mechanisms, fault prevention and detection, and a network monitoring service. Regarding the user interface specifications, the work of the GGF and EGEE project will provide valuable input.
Dynamic Allocation of Resources in VIOLA Christoph Barz, Wolfgang Moll, Markus Pilz University of Bonn, Römerstr. 164, 53125 Bonn, Germany 27. April 2005, University of Bonn, Germany
Agenda Advance Reservations Reservations in VIOLA Security Aspects 2
Advance Reservation Number of admitted reservations Customer Planning of the future Reservation Admission without knowing the full picture Overall network performance Provider 3
Reservation Life Static Provisioning / Immediate Reservation Unknown Duration Negotiation t req t conf ~ t begin Usage/Renegotiation Immediate Reservation Known Duration Negotiation Usage/Renegotiation t req t conf ~ t begin t end Advance Reservation Negotiation Intermediate Usage/Renegotiation t conf t bind t req t begin t end t ack 4
Resource Fragmentation Example Scenario capacity Admission not possible resource capacity admitted requests Worst-Case Scenarios capacity immediate reservation capacity advance reservation 5
Coping with Resource Fragmentation Negotiation Intermediate Usage/Renegotiation Flexible Reservations Flexible Path Selection First Fit / Deadline Rerouting of Planned Flows capacity shift start 1 Malleable Reservations capacity 2 increase capacity reduce duration capacity reduce capacity increase duration Rerouting of intermediate phase flows is inexpensive. Can be done with standard reservation interface! 6
Reservation Types and Parameters reservations static dynamic immediate in advance unknown duration known duration fixed first fit / deadline malleable Advance Reservation Parameters fixed first fit/deadline malleable start stop bandwidth /deadline duration bandwidth start deadline min. bandwidth max. bandwidth amount of data 7
Reservation Dependencies VIOLA Grid Application Reservations Asymmetric Communication Requirements May involve more than two Clusters May be divided into multiple s L2VPN Service Full Mesh of unidirectional MPLS Reservations + Switching Service Symmetric or asymmetric Bandwidth requirements Site A MPLS Site B Site C Decomposition in in multiple (unidirectional) point-to-point Reservations Admission Decision only only for for Bundle Bundle of of Reservations Pre-emption Decision only only for for whole whole Bundle Bundle Failure Failure Recovery: not not correctable correctable error error -> -> cancel/retry cancel/retry whole whole Bundle Bundle Rerouting Rerouting for for point point to to point point Reservations Reservations Layer Layer Mechanisms Mechanisms for for point-to-point point-to-point Reservations Reservations 8
Reservations in the VIOLA Reservations and the Meta-Scheduler Meta-Scheduler Overlay Strategy Yes. Is capacity available from 9:00 to 5:00? Reservation Manager Overlay Strategy: information hiding or minimal information exchange local view and decisions lowers implementation complexity Peer/Integrated Strategy: information distribution or maximal information exchange global view and decisions highest optimization potential 9
Information Strategies in Layered s Interaction with Resources Reservation Manager Is capacity available in MPLS network? Which strategies are possible? MPLS Layer Is capacity available in GMPLS network? GMPLS Layer (SDH) 1. Query MPLS layer for resources. Not enough capacity available! 2. Query GMPLS layer for resources. Where should capacity be added? When should capacity be added? MPLS capacity GMPLS capacity Not enough Capacity in the MPLS layer! Unused Capacity in the GMPLS layer. 10
Strategy for VIOLA In order to query for GMPLS resources, potential MPLS paths must be known to identify connection points to the GMPLS layer! In advance reservations in the GMPLS plane are handled by the reservation manager and a future topology needs to be derived for the MPLS plane. From 9:00 to 5:00 9 Gbit/s are available! VIOLA starts with a single layer approach for MPLS and heads to an integrated layer approach for MPLS and GMPLS. 11
Entities in the Reservation Manager Is capacity available from 9:00 to 5:00? Admission Admission Control: Control: accept/reject accept/reject composed composed reservations reservations trigger trigger rerouting rerouting in in case case of of failure failure Topology Database Reservation Database Path Path Computation: Computation: construct construct future future network network topology topology compute compute a a single single path path for for a a reservation reservation Signaling: Signaling: issue issue commands commands to to network network devices. devices. report report network network status. status. 12
Issues for Larger s Larger networks may contain: multiple domains additional layers (e.g. MPLS, GMPLS, dynamic WDM) multiple information strategies (overlay, augmented and peer model) Focus is on an integrated, intra-domain view for MPLS and GMPLS! Issues could be addressed by connecting multiple network reservation systems. hierarchical mode decentralized mode 13
-based ing ITU TMN Reference Model IETF Model Business Management Level Requirements, Objectives Business Service Management Level Management Level Element Management Level Element SLA Topology Roles of each interface Device- and Technology- Independent View Device-Independent, Technology-Specific View Device- and Technology- Specific View Device specific (MIB, PIB, CLI) QPIM QDDIM Functionality for traffic conditioning and scheduling (DiffServ, RSVP) DiffServ Services for Classifier, Meter, Marker, Queuing, Dropper, Scheduler Automatic Code Generation This turns out to be too complex! We need some form of manual translation from policy to device configuration 14
Resource Manager Architecture User Interface Generic AAA Toolkit (RFC2903) Language Path Computation Resource Manager AAA Rule Based Engine Stateful Sessions Signaling and policing functionality Controlled Resources Topology Service 15
Resource Manager Interfaces Virtual Organization User Interface Resource Manager AAA Controlled Resources Topology Stateful Sessions SLA Local Users Resource Delegation Topology Monitoring Real Organization Service 16
Resource Manager Hierarchy RFC 2903 Decision Point Request Rule Based Engine Application Specific Module Enforcement Point Decision AAA SLA Controlled Resources Path Computation Rule Based Engine Topology Stateful Sessions AAA Path Computation AAA Path Computation Rule Based Engine Controlled Resources Rule Based Engine Controlled Resources Monitoring Topology Monitoring Topology Control Plane Control Plane Element Element 17
Conclusion Advance Reservation Concepts Problem of Resource Fragmentation Flexible Reservations Flexible Path Selection Bundled Reservations Reservations in the VIOLA Resource Manager hides network complexity MPLS and GMPLS layer must be coordinated Focus on an integrated, intra-domain reservation system Security Aspects -based ing Resource Manager Architecture & Hierarchy 18