100 likes | 269 Views
Assured Forwarding (AF): Extend the idea of RIO scheme Service profile specifies the expected capacity Boundary nodes mark packet as in or out of profiles During congestion, out packets are drop first. The network should be provisioned to have sufficient capacity for in packets.
E N D
Assured Forwarding (AF): • Extend the idea of RIO scheme • Service profile specifies the expected capacity • Boundary nodes mark packet as in or out of profiles • During congestion, out packets are drop first. • The network should be provisioned to have sufficient capacity for in packets. • Four forwarding classes, each with three drop precedences. • Each class have a minimum amount of buffers and bandwidth. • A DS node should implement all four AF class, at least two of the three drop priorities. • The resources for each class should be configurable.
Assured Forwarding (AF): • AF PHB codepoints class 1 class 2 class 3 class 4 LOW 001010 010010 011010 100010 Medium 001100 010100 011100 100100 High 001110 010110 011110 100110 • Implementation guideline • A bandwidth partition between classes (WFQ) and drop priotities with a class • E.g: minimum bandwidth 2Mbps, 4 Mbps, 8 Mbps, 16 Mbps, Link would need 30Mbps, four queues with weights 1,2 , 4, 8.
Implementation guideline • Properties for the drop mechanisms • Minimize long-term congestion while allowing short-term fluctuation • Drop mechanism should drop packets from flows of the same long-term characteristics with equal probability. • Drop rate must be proportional to the flow;s percentage of the total traffic • Gradual discard • Relation between AF group and other groups • Not specified • Relationship be configurable and • Any PHB groups that can preempt any AF PHG group • How excessive resources are allocated
AF Service example • Need provisioning • Expected bandwidth service (RIO) • Two Codepoint with one for the four AF classes. • Several services with different provisioning level • Map different services to different class of PHBs
Expedited Forwarding (EF) • Low loss, low latency and assured bandwidth service. • A forwarding treatment for a traffic aggregate where the departure rate of the aggregate’s packet from any DS node must equal or exceed a configurable rate. • Can preempt other traffic • Simple priority queuing • Codepoint 101110 • Implementation guideline • Simple priority queue with a token bucket.
Packet classification • Divides an incoming packet stream into multiple groups. • Behavior aggregate (BA) or multifield (MF) • BA classifies packets based on the DSCP field. • MF uses one or more of the five-tuple • Can support more complex policies, marking packets based on application types, marking packets based on particular source and destination address, etc. • Multidimensional range matching • Requirement: speed, scalability • Algorithms: geometric approach and trie approach
Traffic policing • Metering and marking • Use token bucket. • Dual token bucket algorithm • Two regulators P and C, P is regulated by peak information rate (PIR) and peak burst size (PBS), C is regulated by committed information rate (CIR) and commited burst size (CBS) • Green if a packet passes the tests of P and C • Yellow if a packet passes P but not C • Read if a packet failed both tests
PIR: Peak information Rate PBS: Peak Burst size CIR: Committed Information Rate CBS: Committed Burst Size pkt_size: packet size pkt_arrival: arrival time of previous packet p_tk: the token counter for bucket P c_tk: the token counter for bucket C Token calculation: elapse_time = time_now() – ptk_arrival; p_tk = min[(p_tk+elapsed_time * PIR), PBS]; c_tk = min[(c_tk+elapsed_time * CIS), CBS]; pkt_arrival = Time_now()
If p_tk < pkt_size then mark the packet as RED Else if c_tk < pkt_size then mark the packet as YELLOW p_tk = p_tk – pkt_size; else mark the packet as GREEN p_tk = p_tk – pkt_size; c_tk = c_tk – pkt_size; End if