Quality of Service Issues in MultiService Wireless Internet

Quality of Service Issues in Multi-Service Wireless Internet Links George Xylomenos and George C. Polyzos Department of Informatics Athens University of Economics and Business Athens 10434, Greece polyzos@aueb. gr http: //dias. aueb. gr/~gcp/ Tel. : +30 -1 -8203 -650, Fax: +30 -1 -8203 -325

Outline l Motivation u u l Flexible Link Layer Internet Protocol (FLLIP) u u u l Wireless systems and Internet applications TCP throughput degradation UDP and real-time application issues Proposed approaches Multi-protocol, adaptive, Qo. S/DS aware solution Goals, architecture, implementation Implicit and explicit service selection Conclusion polyzos@aueb. gr 2

Modern Wireless Systems l Opportunities and issues u Digital wireless systems n n n u l Mobility Internet protocols: designed for networks that were u Wired: low error rate n u l Cellular, PCS, 3 G Wireless LANs LEO/MEO Satellites, LMDS TCP: loss congestion Fixed: no mobility, no handoffs Physical layer solutions u Inflexible: one size fits all polyzos@aueb. gr 3

Internet Applications and Protocols l Conventional data exchange applications u u l Usually TCP based Error intolerance Delay tolerance Jitter intolerance (TCP) Interactive and real-time applications u u Often UDP based (plus RTP) Often multipoint (IP Multicast) Some error tolerance Delay intolerance polyzos@aueb. gr 4

Proposed Approaches l Indirect TCP u l Snoop TCP u l works well only in the direction towards the mobile Modifications to TCP u u u l violation of semantics (not end-to-end anymore) Compatibility: usually both ends need to be updated End-to-end retransmissions for a local problem Not multi-protocol: useless for non TCP applications Conventional link-layer schemes u u Inflexible: one service only Irrelevant for some protocols/applications polyzos@aueb. gr 5

Simulation Experiments TCP/UDP IP LL TCP/UDP 2 Mbps 3 ms PHY Wireless Host A l l LL 10 Mbps 1 ms IP PHY Base Station A LL 2 Mbps 3 ms PHY Base Station B IP LL PHY Wireless Host B One and two wireless link scenarios Exponential intervals between errors u l IP TCP/UDP 0. 8 -5. 9% frame loss rates (1 Kbyte frames) TCP: 100 Mbyte file transfer UDP: Voice activated CBR video (1 Mbps) Each test repeated 30 times polyzos@aueb. gr 6

TCP Performance: Throughput polyzos@aueb. gr 7

UDP Performance: Delay polyzos@aueb. gr 8

Flexible Link Layer Internet Protocol (FLLIP) l Address the problem at its source u l Compatible with Internet protocols & architecture u l Fully or mostly reliable Dynamic adaptation to stream/class mix u l Implicitly or explicitly Per stream or class Qo. S differentiation u l IP and higher layers unchanged Aware of Qo. S requirements u l Local solution to a local problem Variable bandwidth allocation Dynamic adaptation to channel conditions polyzos@aueb. gr 9

FLLIP Architecture l l Multiple link layer modules Packet classifier u u l Per class load measurements u l Incoming bandwidth allocations Service class specific processing u l Protocol, TCP/UDP ports IP To. S, DS field Isolation between services Frame scheduler (SCFQ) u Enforces incoming bandwidth allocations polyzos@aueb. gr 10

SCFQ Frame Scheduler l l Enforces incoming allocations l Protects services l Encourages efficiency Self-clocked fair queueing (SCFQ) l Efficient, simple, fair l One queue per class l Heap sort polyzos@aueb. gr 11

TCP Performance: Throughput with FLLIP polyzos@aueb. gr 12

UDP Performance: Delay with FLLIP polyzos@aueb. gr 13

Service Selection l Implicit Qo. S specification u u l Assigns applications to services Protocol and TCP/UDP port fields No changes to Internet protocols and applications Immediate applicability Explicit Qo. S specification u u Assigns traffic classes to services Qo. S provision n u Qo. S differentiation n u Integrated Services, RSVP Differentiated Services More flexible polyzos@aueb. gr 14

Heuristic Packet Classifier l Implicit Qo. S specification polyzos@aueb. gr 15

Differentiated Services Packet Classifier l l Explicit Qo. S specification Dynamic service selection polyzos@aueb. gr 16

Service Measurement and Mobility Feedback l Service selection u u l Standard metrics Refinement Adaptive applications u Mobility polyzos@aueb. gr 17

Conclusions l l TCP performance severely impacted by wireless losses TCP is not the only concern u u l Link layer enhancements u u l Real-time multimedia over UDP New applications and protocols Fast local recovery Customized to underlying link Wireless links: natural choice to introduce Qo. S support u Differentiated services because u Bandwidth is scarce and expensive u Link performance is variable and unpredictable u polyzos@aueb. gr 18