570 likes | 697 Views
Networking. Kihong Park Assistant Professor Department of Computer Sciences Purdue University. Team. K. Park (CS) D. Yau (CS) W. Szpankowski (CS) N. Shroff (ECE) S. Fahmy (CS). Related Faculty. B. Bhargava (CS) D. Comer (CS) E. Houstis (CS) D. Marinescu (CS) V. Rego (CS). Funding (1).
E N D
Networking Kihong ParkAssistant ProfessorDepartment of Computer SciencesPurdue University
Team • K. Park (CS) • D. Yau (CS) • W. Szpankowski (CS) • N. Shroff (ECE) • S. Fahmy (CS)
Related Faculty • B. Bhargava (CS) • D. Comer (CS) • E. Houstis (CS) • D. Marinescu (CS) • V. Rego (CS)
Funding (1) • Network Scheduling • NSF ANI-9875789 (Park, CAREER) • NSF ANI-9714707 (Park) • End System Scheduling • NSF CCR-9875742 (Yau, CAREER) • NSF ESS-9806741 (Grama, Park, Yau)
Funding (2) • Related • NSF ANI-9729721 (vBNS, Purdue-wide) • NSF NCR-9415491 (Szpankowski) • NSF CCR-9804760 (Szpankowski) • NSF NCR-9624525 (Shroff, CAREER) • NSF ANI-9805441 (Shroff, Schwartz) • Sprint (Park, Spafford)
Labs • Network Systems Lab (Park) • Systems Software & Arch. Lab (Yau) • NEWS (Shroff) • Multimedia Compression (Szpankowski) • BOND, COAST, Internetworking Lab, Multimedia DB, RAID
Components End-to-End QoS • Network Scheduling • Packet scheduling • End-to-end control • End System Scheduling • Real-time OS • Distributed scheduling
Research Goal (1) • Network Environment
Research Goal (2) • Diverse QoS requirements • Shared network resources • facilitate QoS-sensitive sharing of network resources
Research Issues • Diverse user QoS requirements • Bursty traffic • Scalability • Efficiency • Resource contention resolution • Selfishness
State-of-the-Art (1) • Resource Reservation & Admission Control • On-demand leasing of bandwidth • QoS guarantees • GPS routers • IETF Int-Serv (e.g., RSVP)
State-of-the-Art (2) • Drawbacks • Wasteful • self-similar bursty traffic • overprovisioning • Scales poorly with system size • per-flow control • admission control & policing
State-of-the-Art (3) • Overkill for many applications • Soft QoS • Graded service levels • Resource contention resolution • ``Who should get what’’ • Noncooperative environment
State-of-the-Art (4) • IETF Diff-Serv • Assured service (Clark) • 2-bit marking bit RIO gateways • Premium service (Jacobson) • Leaky-bucket with priority gateways • aggregate flow control with weak protection
State-of-the-Art (5) • Drawbacks • ``Differentiated’’ service • not QoS-sensitive • Admission control & policing/shaping • impediment to scalability • Weak complexity trade-off • Lack of resource contention resolution
SBS Approach (1) • Dual Architecture • Guaranteed service (GS) • Reservation-based QoS • Small fraction of most stringent apps • Stratified best-effort service (SBS) • Graded service • Large spectrum of QoS-sensitive apps
SBS Approach (2) • Architectural Features • Compatible with reservation scheme • Common substrate of GPS routers • Amortize cost of overprovisioning for GS • Aggregate flow control • Per-flow QoS control • Zero per-flow state at routers • Distributed protocol
SBS Approach (3) • SBS Router
SBS Approach (4) • SBS Control Path
SBS Approach (5) • Properties • Graded services • Beyond differentiated services • QoS-sensitivity • Scalability • No resource reservation • No admission control • Aggregate flow control • Simple user/network interface
SBS Approach (6) • Resource contention resolution • Selfish user behavior • Stability, optimality, efficiency • Pricing • WAN implementation • User-centric networking-centric mechanism • Simple user/simple network realization • Distributed QoS Control
Distributed QoS Control (1) • Component Services • Guaranteed service (GS) • Stratified best-effort service (SBS) • SBS Protocol • Packet scheduling at routers • End-to-end control
Distributed QoS Control (2) • WAN QoS Control: GS
Distributed QoS Control (3) • WAN QoS Control: SBS + GS
Distributed QoS Control (4) • Self-Optimization Procedure Select service class s.t. • . QoS requirement is satisfied • . Least cost, i.e., minimum resource service class • User optimal • Selfishness emulation
Distributed QoS Control (5) • GPS Switch & Multi-Class QoS
Distributed QoS Control (6) • Service Class-QoS Association Table • Ca <Cb Pa >Pb
System Operation (1) • Service Access Point • Direct user access • Access by DB & Security modules • Export API • (GS/SBS, QoS Requirement Vector)
System Operation (2) • Interface with DB & Security Modules
QoS Parameters & Translation (1) • Network QoS • Traffic characterization • Mean/peak/minimum data rates • Reliable/unreliable transport • QoS requirements • Packet loss rate • Delay • Jitter
QoS Parameters & Translation (2) • Application QoS • QoP & other user requirements • Video frame rate (f/s) • Hit rate • Hit rate jitter • Reliability • Security services
QoS Parameters & Translation (3) • Hit Rate • Probability of timely arrival
QoS Parameters & Translation (4) • QoS Translation
QoS Parameters & Translation (5) • Example (hit rate) if Measured Hit Rate < Target Hit Rate then increase allocated resources else decrease resources • QoS translation & control loop
System Performance (1) • vBNS Benchmarking
System Performance (2) • Traffic Configuration
System Performance (3) • SBS Performance
System Performance (4) • Performance Comparison
End-to-End QoS Control (1) • SBS Exports Graded Services • QoS amplification using end-to-end control
End-to-End QoS Control (2) • Real-Time Traffic • Adaptive forward error-correction • Inject redundancy adaptively (network state) • Maintain invariant target QoS • Optimal control problem • AFEC protocol
End-to-End QoS Control (3) • QoS-Redundancy Curve
End-to-End QoS Control (4) • Real-Time MPEG Video Transport
End-to-End QoS Control (5) • Benchmark Experiments
End-to-End QoS Control (6) • Benchmark Data
End-to-End QoS Control (7) • Performance Measurements
End-to-End QoS Control (8) • Redundancy Trace
End-to-End QoS Control (9) • Self-Similar Traffic Control • Exploit long-range dependence • Multiple time scale congestion control (MTSC) • Throughput (SAC, TCP-MT) • Real-time traffic (AFEC-MT)
End-to-End QoS Control (10) • Multiple time scale congestion control
Proposed Research (1) • Implementation of GS/SBS Architecture • Experimental network • ``Purdue Infobahn’’ • Testbed & substrate for MSI • User/network API • 2-Tier network • Backbone: Cisco 72xx routers • LANs: ATM & Gigabit Ethernet