70 likes | 227 Views
Improving and Implementation of Core-Stateless Fair Queueing (CSFQ). Ling Huang U.C. Berkeley cml.me.berkeley.edu/~hlion. Motivation. Stateless solution in current Internet to: employ simple FIFO queueing. achieve scalable and robust. provide best-effort service.
E N D
Improving and Implementation of Core-Stateless Fair Queueing (CSFQ) Ling Huang U.C. Berkeley cml.me.berkeley.edu/~hlion
Motivation • Stateless solution in current Internet to: • employ simple FIFO queueing. • achieve scalable and robust. • provide best-effort service. • Stateful solution and fair queueing to: • provide powerful, guaranteed and differentiated service. • achieve fair share without end-system cooperation. • protect well-behaved flow from misbehaving flow. Can we achieve the best of two worlds?
Edge router Edge router Core router Core router Edge router Edge router Architecture of SCORE • SCORE architecture and CSFQ: • Different tasks in edge routers and core routers: • edge routers perform per flow management. • core routers perform simple FIFO queueing. • Having packets to carry per flow state. Stateless architecture provides services as stateful network! Fig.1 SCORE architecture
Algorithm of CSFQ • Edge routers put flow state in packet‘s header. • Core routers estimate fair share and drop. incoming packets with probability of • A flow should get bandwidth. • Core router updates fair share as follows: if (A > C) new = old * C / F else new = max (ri), where ri active flows Combining fair share computation and probabilistic dropping to approximate fair queueing!
Limitation of Current Approach • Significant latencies between estimation point and congestion point. • Heavy dependence on the estimation algorithms • Failure to work with congestion avoidance algorithm. • Estimation has big deviation when: • many flows startup simultaneously. • bursty traffic. Fig.2 CSFQ with congestion avoidance algorithm Fig.3 CSFQ in bursty traffic
Possible Improving • Introduce mechanisms to: • remedy the heavy dependence on estimation algorithm. • improve its performance with congestion avoidance algorithm. • Develop new algorithm to do estimation: • more accurately. • more continuously.
Implementation in ALTQ/FreeBSD • Per flow state encoding in and extracting from the header of packets. • Embeding CSFQ in the FreeBSD using ALTQ mechanism. • Algorithm can only use 32-bit interger calculations. Fig.4 Alternate queueing architecture