IETF Differentiated Services

Concerns with Intserv:

 Scalability: signaling, maintaining per-flow router state difficult with large number of
flows
 Flexible Service Models: Intserv has only two classes. Also want “qualitative” service
classes
 “behaves like a wire”
 relative service distinction: Platinum, Gold, Silver

Diffserv approach:
 simple functions in network core, relatively complex functions at edge routers (or hosts)
 Do’t define define service classes, provide functional components to build service classes
Diffserv Architecture

Edge router:
- per-flow traffic management
r marking
scheduling
- marks packets as in-profile
and out-profile
b ..
.
Core router:
- per class traffic management
- buffering and scheduling
based on marking at edge
- preference given to in-profile
packets
- Assured Forwarding
 Edge-router Packet Marking
 profile: pre-negotiated rate A, bucket size B
 packet marking at edge based on per-flow profile

Rate A

User packets

Possible usage of marking:

 class-based marking: packets of different classes marked differently
 intra-class marking: conforming portion of flow marked differently than
non-conforming one
 Classification and Conditioning
 Packet is marked in the Type of Service (TOS) in
IPv4, and Traffic Class in IPv6
 6 bits used for Differentiated Service Code Point
(DSCP) and determine PHB that the packet will
receive
 2 bits are currently unused
 Classification and Conditioning

may be desirable to limit traffic injection rate of

some class:
 user declares traffic profile (eg, rate, burst size)
 traffic metered, shaped if non-conforming
Forwarding (PHB)
 PHB result in a different observable (measurable)
forwarding performance behavior
 PHB does not specify what mechanisms to use to
ensure required PHB performance behavior
 Examples:
 Class A gets x% of outgoing link bandwidth over time
intervals of a specified length
 Class A packets leave first before packets from class B
 Forwarding (PHB)

PHBs being developed:

 Expedited Forwarding: pkt departure rate of a
class equals or exceeds specified rate
 logical link with a minimum guaranteed rate
 Assured Forwarding: 4 classes of traffic
 each guaranteed minimum amount of bandwidth
 each with three drop preference partitions
 Diffserv and MPLS
 Both are WAN QoS mechanisms. While Diffserv is
used for traffic aggregation and provisioning of
differentiated services, MPLS is mainly used for
traffic aggregation and load balancing.
 MPLS
 Originally introduced as a WAN mechanism for
forwarding packets using label switching instead
of the IP address-based routing and provide
differentiated QoS.
 It has found its most use in Traffic Engineering
(TE)
 TE requires that traffic follows specific, possibly
nonoptimal, routes to enable diverse routing, traffic load
balancing, and other means of optimizing network
resources.
 MPLS forces traffic into these routes or Label
Switched Paths (LSPs).
 Routers or LSRs
 In the MPLS network, routers are called label
switching routers (LSR).
 Edge LSRs (also called LERs) provide the interface
between the external IP network and the LSP.
 Core LSRs provide transit services through the MPLS
cloud using the pre-established LSP.
 In a SP network, on the ingress the Edge LSR accepts IP
packets and appends MPLS labels.
 On the egress, an edge LSR terminates the LSP by
removing MPLS labels and resorting to the normal IP
forwarding.
 FEC
 The forward equivalence class (FEC) is a
representation of a group of packets that share
the same requirements for their transport. All
packets in such a group are provided the same
treatment en route to the destination.
 Each LSR builds a table to specify how a packet
must be forwarded. The table, label information
base (LIB) comprises of FEC-to-label bindings.
 Labels and Label Bindings
 A label identifies the path a packet should
traverse
 It is encapsulated in a layer-2 header of the
packet -- special MPLS header (aka shim) includes
a label, an experimental field (Exp), an indicator of
additional labels(S), and Time to live (TTL).
 Receiving router uses the label content to
determine the next hop.
 Label values are of local significance only
pertaining to hops between LSRs.
 Labels are bound to an FEC asa result of some
event or policy
 Label Assignment
 Based on forwarding criteria such as
 destination unicast routing
 traffic engineering
 multicast
 virtual private network
 QoS
 MPLS Signaling
 A signaling protocol performs a variety of
functions such as:
 setting up LSPs traversing specified sequences of LSRs
derived from the constraint-based routing (CR) analysis;
 create the path state in each LSR by performing label
allocation, distribution, and binding;
 reserve resources in each LSR including bandwidth, delay,
and packet loss bounds;
 eassign the network resources as necessary;
 dynamically reroute during network congestion and
failures;
 monitor and maintain explicitly routed LSP state
 CR-LDP
 CR-LDP: LDP using constraint-based routing
 LDP provides a common understanding between LSR peers
of the meaning of labels used to forward traffic between
them
 Message categories:

• Discovery -- sent periodically by LSRs to announce their

presence
• Session -- to establish, maintain, and terminate a session
between two LDP peers
• Advertisement -- to create, change, and delete label
mappings to FECs after a session has been established
• Notification -- to signal and provide advisory info.
 Forward path, hard state with no state refreshes
 RSVP-TE
 Signals between LSRs
 Creates a state for a collection of flows between
the ingress and egress points of a traffic trunk
 An LSP aggregates multiple host-to-host flows and
thus reduces the amount of RSVP states in the
network
 Uses firm state where Path and Resv messages
are periodically refreshed but their volume is
significantly reduced
 QoS Routing
 As defined in RFC 2386, QoS “is a set of service
requirements to be met by the network while transporting a
flow.” A flow is “a packet stream from source to a
destination with an associated QoS.”
 Measurable level of service delivered to network users which
can be characterized by packet loss probability, available
bandwidth, end-to-end delay, etc. Expressed as a Service
Level Agreement(SLA) between network users and service
providers.
 QoS-based routing is defined as “a routing mechanism under
which paths for flows are determined based on some
knowledge of resource availability in the network as well as
the QoS requirement of the flows.” A dynamic routing
scheme with QoS considerations.
 QoS Metrics
 Bandwidth, delay, jitter, cost, loss probability
 three types of metrics: additive, multiplicative,
concave
 Let m(n1,n2) be a metric for link(n1, n2). For any
path P = (n1, n2, .., ni, nj), metrci m is:
 additive, if m(P) = m(n1,n2) + m(n2,n3) +…..+ m(ni,nj)
(examples are dealy, jitter, cost, hop-count)
 multiplicative, if m(P) = m(n1,n2) * m(n2,n3) *…* m(ni,nj)
(example is reliability, in which case 0<=m(ni,nj)<=1)
 concave, if m(P) = min{m(n1,n2), m(n2,n3), …, m(ni,nj)}
(example is bandwidth meaning that the bandwidth of the
path as a whole is determined by the link with the
minimum available bandwidth)
 Objectives
 To meet QoS requirements of end users.
 To optimize network resource usage
 to gracefully degrade network performance under
heavy load
 Design Issues(1)
 IP routing protocols such as OSPF, RIP, and BGP
are called “best-effort” routing protocols. They
use only the shortest path to the destination --
single objective optimization algorithms which
consider only one metric (like hop-count).
 Much more difficult to design and implement than
Best-effort routing. Many tradeoffs have to be
made. In most cases the goal is not to find the
best solution but to find a viable solution with
acceptable cost.
 Design Issues(2)
 Metrics and path computation
 how do we measure and collect network state
information?
 how do we compute routes based on the information
collected?
 Mapping of QoS requirements to well defined QoS
Metrics
 Computation complexity associated with path
computation (much of QoS routing based on
multiple constraint optimization is NP-complete).
Many heuristic algorithms exist.
 Design Issues (3)
 Path computation is followed by resource
reservation which means that when the path is
chosen the network state in terms of available
resources is changed and such information needs to
propagated throughout the network.
 Knowledge propagation and Maintenance
 how often the routing information is exchanged between
the routers?
 The tradeoff here is between information accuracy and
efficiency.
 For instance, what is available bandwidth? Is it what is left
after reservation or the actual physically available?
 How do we maintain the info collected?(on demand path
computation, aggregation, routing tables?)
 Design Issues (4)
 Scaling by hierarchical aggregation
 Imprecise state information model. Sources of
inaccuracy:
 network dynamics
 aggregation of routing information
 hidden information
 approximate calculation
 Administrative control -- flow priorities and
preemption, resource control and fairness
 Integrate QoS-based routing and Best-effort
routing
 Intra-domain Vs. Inter-domain
 Dynamic path computation to statically provisioned
paths for a few service classes for intra-domain
 Some common features for intra-domain:
 admission control, optimal resource usage, failure
notices, support for best-effort flows, support for
multicast routing with receiver heterogeneity and shared
reservation styles
 Inter-domain routing scheme have to be scalable
and therefore, simple.
 Cannot be based on highly dynamic network state info
 info exchange between domains should be relatively
static
 Routing Strategies
 Source routing
 distributed routing
 hierarchical routing
 they are classified based on the way the state
information is maintained and the search foe
feasible path is carried out
 Source Routing
 Each node maintains the complete global state,
including the network topology and the state
information of every link
 Based on the global state, a feasible path is locally
computed at the source node
 A control message is sent out along the selected
path to inform the intermediate nodes of their
precedent and successive nodes
 A link state protocol is used to update the global
state at every node
 Source Routing (2)
 Strengths: simplicity through centralization;
avoids many of the distributed computing
problems; guarantees loop-free routes;
conceptually simple, easy to implement, evaluate,
debug and upgrade; centralized heuristics are
much easier to design for some NP-complete
routing problems.
 Weaknesses: communication overhead to maintain
global state; imprecision global state info; high
computation overhead at the source; In short,
source routing has scalability problem.
 Distributed Routing
 Path is computed by a distributed computation
 Control messages are exchanged among nodes and
state information kept at each node is collectively
used for path search
 Requires a distance-vector protocol or link-state
protocol to maintain a global state in the form of
distance vectors at every node. Based on the
distance vectors, the routing is done on a hop-by-
hop basis.
 Distributed Routing (2)
 Strengths: path computation is distributed and
result in shorter routing response time; scalable;
searching multiple paths in parallel for a feasible
path; routing decision and optimization is done
entirely based on local states;
 Weaknesses: dependence on global state; flooding
based algorithms which do not maintain global state
have higher communication overheads; difficult to
design efficient heuristics in the absence of
detailed topology or link-state info; presence of
loops due to inaccurate global state info at
individual nodes (easily detected but alternate
paths are difficult to find)
 Hierarchical Routing
 Nodes are clustered into groups which may be
clustered into higher level groups recursively
creating a multi-level hierarchy.
 Each physical node maintains an aggregated global
state -- contains the detailed state info about the
nodes in the same group and aggregated state info
about other groups.
 Source routing is used to find a feasible path.
 A control message is sent along this path to
establish the connection. A border node in a group
represented by a logical node receives the message
and uses source routing to extend the path through
the group.
 Hierarchical Routing (2)
 Strengths: Scales well; retains many advantages
of source routing as well as distributed routing.
 Weaknesses: aggregated network state introduces
additional imprecision; gets more complicated
when multiple QoS constraints are involved.
 QoS Routing Algorithms
 For Unicast, the problem is to find a network Path
that meets the requirement of a connection
between two end users
 For multicast, the problem is to find a multicast
tree rooted at the sender and the tree covers all
receivers with every internal path from the
sender to a receiver satisfying the requirement
 QoS requirement as a set of constraints
 link constraint (concave metrics)
 path constraint (additive and multiplicative metrics)
 tree constraint
 Algorithms
 Feasible path is one that has sufficient residual
resources to satisfy the QoS constraints of a
connection
 In addition to a feasible path, we also want to
optimize resource utilization -- measured by an
abstract metric cost
 Cost could be in dollars or a function of the
buffer or b/w utilization. Cost of a path is the
total cost of all links on the path
 the optimization problem is to find the least-cost
path among all feasible paths.
 Difficulties
 Diverse applications and different QoS
requirements. Multiple constraints often make the
routing problem intractable -- finding a path with
two independent path constraints is NP-complete.
 Difficult to determine the optimal operating point
for both QoS and Best effort traffic if their
distributions are different. Best-effort traffic
will suffer if overall traffic distribution is
misjudged
 Maintaining up-to-date network state as it changes
dynamically due to transient load fluctuation,
connections in and out and links up and down.
 Graph-based Models
 A network modeled as a graph <V, E>. Nodes (V)
represent switches, routers, and hosts. Edges (E)
represent communication links. Symmetric or
asymmetric links.
 Link state may be a triple consisting of residual
b/w, delay, cost
 Node state can be combined into the state of the
adjacent links
 The delay of a link consists of the link propagation
delay and queueing delay at the node. The cost of
alink is determined by the total resource
consumption at the link and the node.
 State Information
 Local state: each node is assumed to maintain its
up-to-date local state including all delays, residual
b/w on the outgoing links, and the availability of
other resources
 Global state: The combination of the local states of
all nodes. Every node is able to maintain the global
state by either a link-state protocol or a distance-
vector protocol which exchanges the local states
among the nodes periodically.
 Link state protocols broadcast the local state of
every node to every other node. Distance vector
protocols periodically exchange distance vectors
among adjacent nodes. Figures 1 and 2
 Aggregate global state
 Figure 3
 Links and paths
 For some metrics, the state of a path is
determined by the state of the bottleneck link
 link optimization routing -- find a path that has the
largest bandwidth on the bottleneck link -- widest path
 link-constrained routing -- find a path whose bottle neck
bandwidth is above a required value (reduced to link
optimization problem after pruning)
 for some other metrics, the state of the path is
determined by the combined state over all links on
the path
 path optimization -- least cost routing
 path constrained -- delay constrained
 NP-Complete problem classes
 PCPO -- delay-constrained least-cost routing find
the least cost path with bonded delay
 MPC -- delay-delayjitter constrained routing find a
path with both bounded delay and bounded delay
jitter
 These two classes are NP-complete if the QoS
metrics are independent and if they are allowed to
be real numbers or unbounded integer numbers.
 Solvable in polynomial time if all but one metric
take bounded integers; Also if all metrics are
dependent on a common metric (ex. worst-case
delay and delay jitter are functions of b/w in WFQ)
 Chen-Nahrstedt
 Heuristic for multi-path constrained routing
problem. Example: delay-cost constrained
 map the cost (or delay) of every link from an
unbounded real number to a bounded integer;
Solvable in polynomial time
 Source Routing Algorithms
 Maintain a global state at every node
 most algorithms transform the routing problem to
a shortest path problem and then solve it by
Dijkstra’s or Bellman-Ford algorithm.
 Salama et. al. Algorithm
 Distributed heuristic algorithm for delay-
constrained least cost routing problem.
 A cost vector and a delay vector are maintained at
every node by a distance vector protocol
 The cost(delay) vector contains for every
destination the next node on the least-cost (least-
delay) path.
 A control message is sent from the source toward
the destination to construct a delay-constrained
path. Loops may occur and detected if the control
message visits a node twice. Routing process is rolled
back until reaching a node from which the least-cost
path was followed.
 Sun-Landgendorfer
 Improves worst-case performance of Salama et.
al. by avoiding loops instead of detecting and
removing loops.
 A control message is sent to construct the path
 The message travels along the least-delay path
until reaching a node from which the delay of the
least-cost path violates the delay constraint.
 PNNI and QOSPF
 Hierarchical link-state routing protocol
 Topology information is flooded through the
network -- change (LSA)propagated based on a
threshold model
 Traffic classes may be defined to indicate
network resource requirements
 Widest-shortest path (which is a minimum hop
count path with maximum bandwidth) may be pre-
computed for every possible destination.