– Low (bounded) delay, jitter– No loss (if arrivals satisfy specifications)– Minimum departure rate R at routers– (Almost) empty queues for all EF traffic
Provided that arrivals rate < min departure rate Ensured by border routers (aka Boundary conditioners)
How to bound delay, jitter? Define R?– Timescale at which to measure R
Or: a bound E on the `error` (delay on top of Rl) Since R holds only for `sufficiently long periods`…
– Cannot guarantee departure rate if queue is empty Arrival specifications to avoid loss…
Chapter 18 Protocols for QoS Support15
EF PHB: Arrivals SpecificationsEF PHB: Arrivals Specifications EF PHB may include arrival specifications No (congestion) loss if total arrivals in spec
– Total arrivals from all inputs for specific out interface May use Token Bucket Conformance
Parameters:– Token rate R in Bytes/second – Bucket size (or `depth`) B in Bytes– Peak traffic rate p in Bytes/sec– Minimum policed unit length m– Maximum packet size M
Chapter 18 Protocols for QoS Support18
EF PHB: Per-Packet Delay SpecEF PHB: Per-Packet Delay Spec EF-able router MUST satisfy: dj,I
P≤ fj,IP + EI
P
– dj,IP: time when router finished sending the jth packet
Received for I
– With specified rate RI and error EIP for interface I
– EIP caps the delay error on interface I (for rate RI)
Where: fj,IP=max{aj,I , min(dj-1,I
P, fj-1,IP)}+lj,I / RI
– is the `ideal` time for finishing to send the jth packet– aj,I is time when router finished receiving jth packet for I– lj,I
is the length of the jth packet received for I f0,I
P=0, d0,IP=0
Spec also has aggregate delay requirements – see RFC– Different if the router is non-FIFO (as many are)
Chapter 18 Protocols for QoS Support19
Bounding Delay and Jitter Bounding Delay and Jitter Given:
– EIP : maximal per-packet delay error on I
– Total inputs towards I conform to token-bucket with rate RI and size BI
Then a bound of delay and jitter is:D=BI / RI + EI
P
If there is a minimal delay Emin<EIP
then we can improve bound on jitter to:J=D-Emin
– How to improve bound on jitter? And what for?
Chapter 18 Protocols for QoS Support20
Per-Domain Behaviour (PDB) and Per-Domain Behaviour (PDB) and Service Level Agreement (SLA)Service Level Agreement (SLA)
– Attributes: throughput, loss, latency, jitter under specified conditions
– Network parameters: max number of hops, min bandwidth, max delay,…
– Allow composition of domains into DS regions SLA: Between provider (of DS domain) and customer
– Customer: user organization or other DS domain– SLS: Service Level Specifications (parameters of SLA)– Customer visible, quantified, end-to-end QoS – Provider can use PDB to meet SLS
– Book chapter is also in http://www.isoc.org/pubs/int/cisco-1-5.html– MPLS IETF WG, MPLS forum, MPLS Resource Center, …
Mid-1990s: Efforts to marry IP and ATM– Motivation: ATM switches were much faster than routers– IP switching (Ipsilon), Tag switching (Cisco), …
Routing (e.g. OSPF) define path between end points Diff-Serv: assign packets to class on entering Net MPLS: assign a fixed path for each flow
– Simpler, faster routing/switching of packets– Connection-oriented QoS support (cf. DiffServ)– Traffic engineering: choose and change path for each flow– Virtual private networks– Multi-Protocol support
– Forward packets based on MPLS label, not on IP header– Add / swap / merge labels (more later)
Labels identifies flow of packets– Between end points or multicast destinations– Flow = FEC (Forward Equivalency Class)– Flow defined by src/dest IP addr, port; protocol; DSCP
Connection oriented: each flow has…– Specific path through LSRs– Specific QoS parms: resources, queue/discard policy– Path determined by routing protocol e.g. OSPF– Labels assigned manually or by proper protocol
Enhanced RSVP or LDP (Label Distribution Protocol)
Chapter 18 Protocols for QoS Support25
MPLS Domain OperationMPLS Domain Operation
Chapter 18 Protocols for QoS Support31
Explanation - SetupExplanation - Setup
Labelled Switched Path (LSP) route selection
QoS parameters established along LSP:– Resource commitment– Queuing and discard policy at LSR (PHB)
Labels assignedLocal significance onlyManually or using protocol
Chapter 18 Protocols for QoS Support32
Route SelectionRoute Selection
Selection of LSP for particular flow Hop-by-hop
– LSR independently chooses next hop– Ordinary routing protocols e.g. OSPF– Doesn’t support traffic engineering or policy routing
Explicit– LSR (usually ingress or egress) specifies some or all
LSRs in LSP for given flow– Allows traffic engineering and policy routing– Selected by configuration, or dynamically
Chapter 18 Protocols for QoS Support33
Constraint Based RoutingConstraint Based Routing
Take into account traffic requirements of flows and resources available along hops
Current utilization, existing capacity, committed services
Additional metrics over and above traditional routing protocols (e.g. OSPF, BGP)– Maximum link data rate– Current capacity reservation– Packet loss ratio– Link propagation delay
Chapter 18 Protocols for QoS Support34
Label Distribution and Resource Label Distribution and Resource Allocation Allocation Setting up LSP for a flow… each LSR:Assign in-label to incoming packetsInform all upstream LSRs of in-labelReceive out-label from downstream LSRManually or by label-setup protocol
