QoS support over Stateless DiffServ Networks by means of localized measurements and decisions

1. University of Perugia QoS support over Stateless DiffServ Networksby means of localized measurements and decisions Nicola Blefari-Melazzi University of Perugia, Italy blefari@diei.unipg.it • Statistica e telecomunicazioni: nuove metodologie per nuovi problemi • 3/7/01, CNR, Sede Centrale– Roma

2. A A 3 2 1 3 2 1 A B 3 2 1 3 2 1 C C 3 2 1 3 2 1 C 3 2 1 QoS support: background Circuit-switched networks, ATM, IntServ guaranteed performance Refused Admission control

3. A A A 3 2 1 3 1 3 2 1 B B 3 2 1 B 3 2 1 1 3 2 C C 3 2 1 3 2 1 C 3 2 QoS support: background Today’s Internet: no admission control best effort performance dropped/delayed No admission control

4. QoS support: background • Integrated Services approach (IntServ) • Traffic Descriptor Declaration, CAC, Policing • RSVP: explicit resource reservation along the signaling path • “hard” QoS guarantees • Per-flow management->complexity, limited scalability (ex. ADPCM coding requires 32 kb/s for a voice channel. A single OC-12 link (622 Mb/s) can support up to 20000 flows) • Differentiated Services approach (DiffServ) • Edge-only state implies that the service indication must be carried in the packet->establishing different “priority levels” • simplicity and scalability but no QoS guarantees • the lack of an admission control scheme means that, upon overload in a given service class, all flows in that class suffer a degradation of service • “actual” DiffServ appears to be a way to provide "better-than-best-effort" service, i.e. an intermediate solution between the extremes of hard guarantees and the vagaries of best effort

5. QoS support: recent RFCs • RFC 2998 and RFC2990 (Nov. 2000) recognize that: • “both the IntServ architecture and the DiffServ architecture have some critical elements in terms of their current definition which appear to be acting as deterrents to widespread deployment • there appears to be no single comprehensive service environment that possesses both service accuracy and scaling properties • further refinement of the QoS architecture is required to integrate DiffServ network services into an end-to-end service delivery model with the associated task of resource reservation • it is then suggested to define an admission control function which can determine whether to admit a service differentiated flow along a nominated network path”

6. Measurement-based Admission Control • MBAC: each router takes admission control decision based on run-time measures of aggregate traffic handled (necessity of signaling) Measur.& Decision Measur.& Decision Measur.& Decision Destination Source Data Data Signalling Signalling Signalling Signalling

7. Ack Measur. Core Router Rin Rout 2 Ack Source Probing Destination Probing 1 Data 3 3 Data Endpoint Admission Control • EAC: admission control managed by pure end-to-end operation, by means of a probing phase (long set-up time or imprecise measurements)

8. Measur.&Decision ->Gate open or closed 3 2 1 Destination Source Ack Ack Ack Ack Probing Probing Probing Probing Data Data GRIP basic principles • SOURCE NODE: at connection set-up starts a probing phase • injection of a “probe” packet (tagged as such) • activation of a timer (DECISION_WAITING state) • if an ACK is received from the destination before timeout expiration, it can start emitting “data” packets (tagged as such) • DESTINATION NODE:starts a DECISION_ACTION procedure, after receiving a probe packet; if it is willing to accept the connection, it replies with an ack packet • ROUTERS: drop probes (GATE closed) when congestion is estimated (Gauge)

9. GRIP: Internal Router Decision Criterion • The MBAC module measures the aggregate traffic and implements a Decision Criterion • The Decision Criterion acts as a switch on the probe queue • ACCEPT state: probe packets served • REJECT state: probe packets discarded

10. Ack R1 R1 Probing packets HP HP LP LP GRIP’s operation A t0+ t <TO B t0 t0+t Measurament & decision

11. GRIP: end-to-end operation • Probing and data packets marked by different DSCPs • The decision criterion (DC) must not be necessarily based on traffic measurements but could be independently defined by network operators (i.e., by implementing a separate probing queue with given capacity and buffer space, to suitably limit accepted probe packets, or by relying on lower layers capabilities, e.g., ATM, MAC) • The notion of internal router congestion is not standardized, and it is up to each specific router DC implementation to locally determine if, and when, congestion arises • The internal (arbitrarily sophisticated and performing) router decision is summarized in the router state (ACCEPT vs. REJECT)

12. GRIP: end-to-end operation • The router state is not notified to the end points by means of explicit signaling information transmission • end points rely on probing packet losses (i.e. dropped by routers in the REJECT state) as an implicit signaling pipe, of which the network remains unaware • when the router is in the ACCEPT state, it advertises that it can admit new connections by allowing probing packets to be served. Conversely, when the router is in the REJECT state, no probing packets are forwarded • locally blocking probing packets implies aborting all concurrent setup attempts of connections whose path crosses the considered router. Conversely, a connection is successfully setup when all the routers crossed by a probing packet are found in the ACCEPT state

13. GRIP: end-to-end operation • no explicit agreement among entities (e.g., probing packet format, probing/feedback packets payload contents) is necessary to run GRIP • this principle can be the way to provide a smooth migration path, consisting in distributed admission control schemes of increasing complexity and effectiveness, which can indeed operate over a multi-provider and multi-vendor Internet • each GRIP component accounts for an extremely broad class of possible implementations, and independent implementations of different components can inter-operate

14. The Migration Path • GRIP can seamlessly operate, at the expense of perceived performance, even in the legacy Internet and in a DiffServ network • (failed reception of probing packets means that a congestion condition occurs in the network, as in TCP) • in these conditions, GRIP provides at least a form of stability that impedes persistent link congestion and allows a receiver capability negotiation • The performance of GRIP are related to the capability of routers to locally take decisions • Thus, GRIP opens up a future smooth migration path toward gradually improved QoS, as routers in different domain will be enhanced, without losing inter-operability with installed devices • Strict end-to-end QoS guarantees are eventually provided when all the crossed routers are equipped with GRIP capabilities

15. GRIP respects the driving concepts: • Backward compatibility (with legacy Internet) • Smooth migration path • toward a QoS capable infrastructure by means of independent upgrades • Scalability: • no state information • handling of traffic aggregate • Distributed operations: • all network devices operating autonomously • the exchange of signaling is implicit • inter-working among different sub-networks is simplified • Performance calibration: • "tuning knobs"that allows independent network operators to set target performance levels • Compatibility with AF PHB (RFC2597)

16. In(t) Out(t) 1 Source BTS PS DLB rS Preliminary performance evaluation • Assumptions: • We consider a full-fledged QoS GRIP domain • The traffic is regulated at the network edge by means of DLB devices • the sources are greedy (they use all the emitting opportunity they have)

17. Estimation of Admitted Sources • Conservative estimation of the number of active sources, N, based on the worst case behavior of the DLB output processes • Maximum number of admissible sources, K, determined off-line by means of acceptance rules based on loss/delay constraints • New requests are accepted as long as N≤K • K is a “tunable knob” and can be chosen by the network operator by trading network utilization with user-perceived performance (e.g. loss, delay, etc) • The performance figures can then be enforced by a suitable choice of K. The operator chooses the target levels; the latter are mapped in a value of K and GRIP enforces such value

18. Inaccuracy of the Estimation Process, which is Based on a Worst Case Estimation

19. Numerical results: sample path

20. Numerical results: Effect of window size

21. Numerical results: effect of external loading

22. Transient Management • when a new flow is admitted, there exists a transient time in which its presence is not perceived by the measurements. Solution: activation of a “stack” variable to keep memory of the amount of “transient” flows

23. 90 80 70 60 Number of admitted flows 50 40 30 Probe loss: 0, 0.25, 0.50, 0.75 20 10 0 0 100 200 300 400 500 600 Simulation time (seconds) Multinode scenario • Performance degradation induced by the stack implementation almost negligible

24. Mobility support: open issues • Micromobility support • Mobility is handled hierarchically, to reduce traffic generated by handovers and the relevant latency (a local “anchor” node handles mobility, without involving Home Agent and Correspondent Node) • Integration between Mobility and QoS • QoS during and after the handover (QoS-Aware Handover)

25. Correspondent Node Internet Local Mobility Gateway Regional HO Local Mobility Network Plain MIP HO Access Router “B” Access Router “A” Mobile Node Local Handover Global Handover Micromobility support

26. QoS aware Handover • Extensions needed to support QoS-Aware handovers: • Perform a test of the availability of the resources on the new path before switching the QoS-sessions of the Mobile Node on that path • Maintaining the current path active while testing resources availability on the new path • Avoid to reserve resources both for the new and the current path along the entire end-to-end path • Manage the mobility of the Mobile Node in such a way that the Handover is transparent to the correspondent node

27. Correspondent Node Internet Local SHA R6 R5 1 2 R1 R2 R3 R4 Route Update MN MN 1 2 QoS aware Handover in micromobility • The SHA can be introduced in IP Micromobility Architectures such as • Hierarchical Mobile IP • Cellular IP • Regionalized Registration • Once again, IntServ (with per flow reservation) seems inappropriate: • necessity of handling states, whose number is now doubled • RSVP has not been designed to handle two contemporary routes • intra-domain user migration cannot be managed in a localized manner (as micromobility management suggests), but requires RSVP signaling to travel back and forth to the destination node

28. Correspondent Node 1 Trigger packet Internet 2 Probing packet 3 Feedback packet Local Mobility Gateway Local Probing Local Mobility Network 1 3 2 Global Probing 2 3 Access Router “B” Access Router “A” 1 Mobile Node Local Handover Global Handover Possible solution: mobile GRIP

29. References • G. Bianchi, N. Blefari-Melazzi: "GRIP: QoS support over Stateless DiffServ Networks by means of localized measurements and decisions", Lecture Notes on Computer Science, Springer-Verlag, volume 1989 (a more detailed technical report can be found at http://drake.diei.unipg.it/netweb/GRIP_tech_rep.pdf) • G. Bianchi, N. Blefari-Melazzi, M. Femminella, F. Pugini: "Performance Evaluation of a Measurement-Based Algorithm for Distributed Admission Control in a DiffServ Framework", to appear on 2001 Tyrrhenian International Workshop on Digital Communications, 17-20 Settembre, 2001, Taormina, Italy (in Lecture Notes on Computer Science, Springer-Verlag, Sergio Palazzo, Ed.). • G. Bianchi, N. Blefari-Melazzi, M. Femminella: “A Migration Path to provide End-to-End QoS over Stateless Networks by Means of a Probing-driven Admission Control”, Internet Draft, draft-bianchi_blefari-end-to-end-QoS-00.txt, work in progress, http://www.ietf.org/ID.html • G. Bianchi, N. Blefari-Melazzi: " Per Flow Admission Control over AF PHB Classes”, Internet Draft, draft-bianchi_blefari-Adm-Contr-over-AF-PHB-00.txt , work in progress, http://www.ietf.org/ID.html • G. Bianchi, N. Blefari-Melazzi, M. Femminella, F. Pugini: “Stateless Per-flow Admission Control in a Mobile IP Environment”, http://drake.diei.unipg.it/netweb/GRIP_mobile.pdf