Transcript QoS in Converged Networks - School of Information Technology

QoS in Converged Networks
ITK 477
QoS in PSTN
• In traditional telephony, quality of service for
each and every phone call is guaranteed by the
constant availability of dedicated bandwidth.
• Most digitally encoded call paths on the PSTN use
the same codec, G.711, so transcoding isn’t
necessary.
• Almost no processing bottlenecks will be found
on the PSTN, and since the system isn’t generally
packet-based, there is almost never degradation
in perceived call quality as a result of congestion.
QoS in Packet Networks
• When bandwidth availability drops, as more
packets are sent on the network, throughput
slows.
• Until a certain breaking point, bandwidth
availability can be compromised while still
allowing data through; the transmission just
slows down.
• Some applications tolerate congestion and slow
• throughput better than others. The more
tolerance an application has, the higher its error
budget is said to be.
Latency
• Slowness of transmission—latency—is the
enemy of multimedia traffic
• Solution to the latency problem: technique
that allows local and end-to-end guarantees of
bandwidth and prioritization of real-time
traffic over less sensitive traffic.
• QoS protocols and standards: 802.1p, 802.1q
VLAN, DiffServ, RSVP, and MPLS.
•
Call-quality scoring
• Mean opinion score (MOS): Listeners hear
sound samples from calls of varying quality,
recorded during different sets of network
conditions.
• Everbody rates the sample’s quality on a scale
of 1 to 5, with 5 being the best quality.
• G.711’s highest perceived quality score is 4.4.
By comparison, G.729A’s is only 3.6.
• See next figure:
Don’t use G.729A across a fast Ethernet link
because the quality perceived by users will be
lower than it ought to be.
Noise
• One of the biggest factors in perceived quality
is noise.
• Additive noise is the unwanted signals that
accompany all transmissions of sound.
Subtractive noise is an interruption or
reduction of the sound transmission, such as
that caused by packet loss.
Noise
• Multimedia traffic, such as VoIP does introduce
new kinds of noise, broadening the traditional
definition to include everything shown in the next
figure.
• While noise cannot be entirely avoided, it should
be minimized.
• One of QoS’s roles is to help us avoid situations in
which poor service at the lower layers of the
network results in additive or subtractive noise.
Noise
Class of Service versus Quality of
Service
Standards
Latency, Packet Loss, and Jitter
• Latency (also called lag) is caused primarily by slow network links.
• End-to-end latency, in the case of VoIP, is the time it takes from the instant
the caller utters something until the time the receiver hears that
utterance.
• Round-trip latency less than 150 ms is not immediately noticeable, but
latency higher than 150 ms is discouraged, and latency higher than 300 ms
is considered unacceptable.
• Latency has the following effects on telephony and video applications:
• Can slow down the human conversation
• Can result in caller and receiver unintentionally interrupting each other
• Can worsen another Quality-of-Service problem: echo
• Can cause synchronization delays in conference-calling applications
The best ways to beat latency are to use low-packet-interval codecs and maintain
fast network links, because QoS protocols alone cannot directly improve
latency’s impact. That is, they can’t speed up your network.
Sources of Latency
• Framing and packetization
• • Software processing and packet loss concealment (PLC; replacing
the sound that would presumably have been produced by a packet
that was lost with sound that is predicted based on the sequence of
packets received before it and (when extensive buffering is used)
after it)
• Jitter buffering
• Routing and firewall traversal
• Transcoding
• Media access and network interfacing
Minimizing latency is an important way to maximize the multimedia
(VOIP) network’s perceived quality of service.
Cont.
• The two biggest sources of latency are
framing/packetization, which can add up to 30
ms of latency, and routing, which can add 5–
50 ms per hop.
• Another big contributor is transcoding (See
next figure)
Transcoding Latency in ms
Packet Loss
• Even with PLC in force, packet loss rates on a
VoIP network should be kept below 1%.
• A drawback of PLC is that it can increase
latency.
• Experimentation with Packet Loss
Concealment (PLC)-equipped codecs should
be done to determine how negative the
latency-impact PLC is in your VoIP network.
Jitter
• It’s the variation in latency time from one packet to the next.
• It causes packets to arrive out of order, leaving gaps in the framing
sequence of the voice/video signal.
• Jitter is at its worst when voice traffic must travel through several
routers on the network.
• Different routers, especially those at ISPs, may be configured to
queue and forward different kinds of traffic in different ways.
• Others may be loadbalancing, which can contribute to jitter.
• The main goal of QoS protocols is to eliminate jitter.
• Devices called jitter buffers, in endpoints and VoIP servers, can
minimize the effect of jitter, too. But, like PLC measures, they do so
by increasing latency.
Class of Service (COS)
• CoS systems work to prioritize traffic on a single data link.
• While QoS refers to the greater network, CoS refers to only
a single data link.
• The key difference is that CoS is a single-link approach,
while QoS is an end-to-end approach.
• Class of Service systems define per-hop behavior, so they
cannot guarantee a service
• level in terms of capacity or speed.
• Two key standards support CoS:
802.1p/ToS
DiffServ
802.1p
• 802.1p uses a 3-bit portion of the Ethernet packet
header to classify each packet into
• a particular level of precedence on the local data link.
• Type of Service (ToS) is the portion of the IP packet
header that stores the same precedence information.
• If your VoIP network will be more than 70% data-tovoice and unlikely to reach capacity, packet
prioritization techniques like LAN-oriented 802.1p and
its WAN cousin DiffServ are adequate.
• The next table lists the suggested, generic service
names.
Suggested 802.1p classes
Differentiated Services (DiffServ).
. When a packet reaches the edge of the
network, either from an endpoint or from a
• remote network, DiffServ tags that packet’s
ToS header based on the priority established
for that packet by policy.
• Once admitted into a DiffServ-equipped WAN,
however, all subsequent router hops must
enforce the priority set by the edge router
that admitted the packet.
Policy servers
• Common Open Policy Service, or COPS, is a way of
storing and querying centralized
• policy information on the network.
• DiffServ can use COPS to obtain its marching orders for
how to handle traffic coming into the network.
• In a COPS scheme, a centralized server called the policy
server contains a policy record of traffic shaping and
prioritization preferences that DiffServ or another
CoS/QoS mechanism can retrieve.
• Another IETF recommendation, LDAP (Lightweight
Directory Access Protocol), can also be used as the
basis of a policy server.
DiffServ Code Points (DSCP)
• DSCP classes are IP packet headers DiffServ
associates with different levels of importance.
• Since they’re 6 bits in length, DSCPs can be used
to define quite a wide scale of possible service
levels. Most implementations support only 3 bits,
replacing the 3 bits in IP’s ToS header.
• DSCP per-hop behaviors break down into three
basic groups, interchangeably called PHB classes,
traffic classes, or DSCP classes:
DSCP Classes
• AF Assured Forwarding, a highly expedient
DSCP class, sometimes used to tag signaling
packets such as H.245/H.225 and SIP packets.
• EF Expedited Forwarding, the most expedient
DSCP class, used to tag packets carrying actual
sound data.
• BE Best Effort, a nonexpedient DSCP class,
used to tag non-voice packets. Many DiffServ
decision points don’t use BE.
802.1q VLAN
•
•
•
•
•
•
Broadcast domain per network segment means that when a packet comes across
the segment destined for a local host whose hardware (MAC) address has not yet
been resolved (ARPed) and associated with a certain switch port on the Ethernet
segment, a broadcast to all ports is done in order to find a host with the right MAC
address that’s supposed to receive the packet.
Once the port with the correct recipient is found, an ARP record is recorded in the
switch so that all future traffic destined for that MAC address can go to that port
rather than being broadcast.
One problem is that the broadcast traffic can be a waste of bandwidth.
Another problem is that, when broadcasts occur, every device on the network can
receive them, which is a potential security hazard.
802.1q VLAN (virtual LAN) is a way to separate Ethernet traffic logically, secure
Ethernet broadcast domains, organize the network by separating network
protocols into their own VLANs
Each VLAN is a logically separate broadcast domain—even if it coexists with other
VLANs on the same physical segment.
Layer 2 Switching
• With most vendors’ Ethernet equipment, to
create VLANs, each switch port is assigned a
VLAN tag—a numeric identifier that is unique
within the network.
• This tag identifies the VLAN in which that port
participates. Once the tag is assigned, the device
connected to that port will receive traffic only
from the assigned VLAN and will be able to send
traffic only to the assigned VLAN.
VLANs
Layer 3 Switching
• Sometimes Ethernet switches can be used to
groom, inspect, or route traffic.
• Layer 3 switching accomplishes some router-like
activities: queuing, routing, and packetinspection.
• It can be used to shape the traffic on the data link
based on each packet’s
• characteristics.
• For example, it’s possible to drop all non-voice
traffic by filtering protocol types (UDP, TCP, etc.)
and port numbers.
Quality of Service
• Intserv (Integrated Services) is an IETF recommendation for provided
dedicated bandwidth to individual flows, or media channels, on an IP
network.
• The media channels are referred to by their sockets
• RSVP (Resource Reservation Protocol) is the recommended signaling
protocol for
• Intserv.
• The purpose of RSVP is to ensure that the network has enough bandwidth
to support
• each call, before any data is passed through the media channel.
• RSVP adds decision-making points to the core network, increasing the
processing overhead requirement on core routers.
• RSVP is the perfect solution for bandwidth allocation over slower links,
because it guarantees availability for each RTP stream, rather than giving a
“best effort.”
Example: Slow Links Between Routers
H.323
1. H.245 negotiates the codec and establishes RTP sockets that will be used on either
end of the media channel. These two sockets—the IP addresses and port
numbers—together form the session ID that RSVP will use to refer to this
RTPsession. RSVP calls the session ID a flow ID.
2. The gateway router for the caller, B, sends a path message (PM) to the next hop, B,
along the way to the remote gateway router, D. This PM will continue to be
forwarded from one hop to the next in order to establish the QoS path.
3. B records the latency added as the PM reaches it, along with minimum latency, jitter
ranges the router is willing to guarantee. Then, the PM is sent to the next router
along the path, in this case, C.
4. C records the latency added as the PM reaches it, along with minimum latency, jitter
ranges the router is willing to guarantee. Then, the PM is sent to the next router
along the path, in this case, D.
5. When the PM reaches the remote gateway router, D, cumulative latency and jitter
are calculated. The result is a profile call the ADSPEC, and the portion of the RSVP
header used to accumulate QoS data during the PM is called the ADSPEC header.
Link delays and maximum jitter
readings are recorded for each hop.
RSVP
• When the remote gateway router reads the
ADSPEC data and makes the determination, it
can do one of two things:
• Give up, resulting in a busy tone for the caller,
or
• Trigger the reserve message (RM) to set up
the traffic contracts with each router in order
to reserve bandwidth for the call.
Reserve Messages (RM)
1. The remote gateway router (D) sends the reserve
message to the previous router in the path. The
sender and receiver RTP sockets are confirmed,
and a contract is established for the timeout
value in seconds, sustained throughput, and peak
throughput required by the RTP session.
2. The previous router in the path (C) sends a
similar RM to its previous router in the path (B).
3. Router B sends router A another RM.
RM Confirmation
1. Router A sends a reserve confirm message to
router B if it agrees to guarantee the bandwidth
and timeout values requested, or a rejection
message if not.
2. Router B sends router C a similar response. If the
first response, from router A, was a rejection,
then all subsequent responses will be rejections
as well.
3. Router C sends router D a similar response. If the
first or second was a rejection, then this response
will be a rejection as well.
RSVP Service Levels
• RSVP defines three service levels in RFC 2211:
• Best Effort
A class of service that has no QoS measures whatsoever. On
Cisco routers, the fair-queuing feature is used to enable
Best Effort service.
• Controlled Load
Allows prioritization of traffic over multiple routers like
DiffServ but includes core routers in the decision-making
process.
• Guaranteed
No packets will be lost, bandwidth will be constant, and delay
will be within the prescribed ranges set up in the traffic
contract.
MPLS
•
•
•
•
•
MPLS bears great similarity to ATM signaling but borrows heavily from RSVP. Unlike
ATM, which incurs a 25% overhead on TCP/IP traffic (called the ATM “cell tax”),
MPLS doesn’t use its own framing format, just its own labeling format.
The purpose of MPLS labels is to identify the paths and priorities associated with
each packet. The paths correspond to the media channel of the VoIP call, while the
priorities respond to the QoS level of service negotiated for those channels, just
like RSVP.
But like DiffServ, MPLS can use a dumb network core. If a packet is carrying a label,
all a router has to do is send it along the labeled path, rather than making a
redundant assessment of the packet’s payload.
MPLS inserts itself partially in layer 2 and partially in layer 3 on the OSI model. Its
frame header sits between the IP header and the Ethernet header on an Ethernet
network or between the label header and the payload on an ATM network.
What’s important to know is this: MPLS resides outside the reach of the network
protocol, like 802.1p. framing protocol (Ethernet framing, for example). This
makes it invisible to the higher layers.
Multiprotocol Label Switching
(Handout)
• Multiprotocol Label Switching (MPLS)
– Born of Cisco’s tag switching, designed with large-scale
WAN in mind, MPLS was proposed by the Internet
Engineering Task Force (IETF) in 1997.
– Core specifications for MPLS were completed by IETF in
the fall of 2000.
– By plotting static paths through an IP network, MPLS
gives service providers the traffic engineering
capability they require while also building a natural
foundation for VPNs.
• Traffic engineering allows service providers to do
two things: control quality of service (QoS) and
optimize network resource utilization.
– MPLS also has the potential to unite IP and optical
switching under one route-provisioning umbrella.
38
How MPLS Works
• “MP” means it is multiprotocol. MPLS is an
encapsulating protocol, it can transport a multitude
of other protocols.
• “LS” indicates that the protocols being transported
are encapsulated with a label that is swapped at each
hop.
– A label is a number that uniquely identifies a set of data
flows on a particular link or within a particular logical link.
– The labels are of local significance only – they must change
as packets follow a path – hence the “switching” part of
MPLS.
39
How MPLS Works
• MPLS can switch a frame from any kind of layer-2 link
to any other kind of layer-2 link without depending
on any particular control protocol.
• ATM can only switch to and from ATM and can use
only ATM signaling protocols, such as PNNI (Private
Network-to-Network Interface) and IISP (Interim
Interface Signaling Protocol).
40
MPLS
• Since IP is a connectionless protocol, it cannot
guarantee that network resources will be
available.
• Additionally, IP sends all traffic between the same
two points over the same route. During busy
periods, therefore, some routes get congested
while others remain underutilized.
– One key difference between MPLS and IP is that
packets sent between two end points can take
different paths, based on different MPLS labels.
• Without explicit control over route assignments,
the provider has no way to steer excess traffic
over less busy routes.
41
MPLS
• MPLS tags or adds a label to IP packets so they
can be steered over the Internet along predefined
routes.
• MPLS also adds a label identifying the type of
traffic, path and destination.
• This allows routers to assign explicit paths to
various classes of traffic.
• Using explicit routes, service providers can
reserve network resources for high-priority or
delay-sensitive flows, distribute traffic to prevent
network hot spots and pre-provision backup
routes for quick recover from outages.
42
MPLS
• An MPLS network is comprised of a mesh of label
switch routers (LSRs)
– LSRs are MPLS-enabled routers and/or MPLS-enabled
ATM switches.
• As each packet enters the network, an ingress LSR
assigns it a label based on its destination, VPN
membership, type-of-service bits, etc.
• At each hop, an LSR uses the label to index a
forwarding table. The forwarding table assigns
each packet a new label, and directs the packet to
an output port. To promote scaling, labels have
only local significance
• As a result, all packet with the same label follow
the same label switched path (LSPs) through the
43
network.
Multiprotocol Label Switching
(
)
Stallings, High-Speed Networks
45
How MPLS Works
• With MPLS you can support all applications on an
IP network without having to run large subsets of
the network with completely different transport
mechanisms, routing protocols, and addressing
plans.
• Offers the advantages of circuit-switching
technology, including bandwidth reservation and
minimized delay variations for voice and video
traffic, plus all the advantages of existing besteffort, hop-by-hop routing.
• Allows service providers to create VPNs with the
flexibility of IP but the QoS of ATM.
46
MPLS Labels
• MPLS supports three different types of label
formats.
– On ATM hardware it uses the well-defined Virtual
Channel Identifier (VCI) and Virtual Path Identifier (VPI)
labels.
– On frame relay hardware, it uses a Data Link
Connection Identifier (DLCI) label.
– Elsewhere, MPLS uses a new, generic label known as a
Shim, which sits between layers 2 and 3.
• Because MPLS allows the creation of new label
formats without requiring change in routing
protocols, extending technology to new optical
transport and switching should be
straightforward.
47
MPLS Label Stacking
• Another powerful attribute of MPLS is Label
Stacking.
• Label stacking allows LSRs (label switched router)
to insert an additional label at the front of each
labeled packet, creating an encapsulated tunnel
that can be shared by multiple LSPs (label
switched paths).
• At the end of the tunnel, another LSR pops the
label stack, revealing the inner label.
• An optimization in which the next-to-last LSR
peels off the outer label is known in IETF
documents as “penultimate hop popping”.
48
MPLS Label Stacking
• ATM has only one level of stacking, virtual channels
inside of virtual paths.
• MPLS supports unlimited stacking.
– An enterprise could use label stacking to aggregate
multiple flows of its own traffic before passing it on to the
access provider
– The access provider could aggregate traffic from multiple
enterprises before handing it to a backbone provider
– The backbone provider could aggregate traffic yet again
before passing it off to a wholesale carrier.
49
MPLS Label Stacking
• Service providers could use label stacking to
merge hundreds of thousands of LSPs into a
relatively small number of backbone tunnels
between points of presence.
• Fewer tunnels means smaller route tables,
making it easier for providers to scale the
network core.
50
MPLS Evolution
• However, the IETF and the MPLS Forum still have
issues to resolve.
– They must reconcile MPLS with DiffServ, so that type-ofservice markings can be transferred from IP headers to
MPLS labels and interpreted by LSRs in a standard manner.
– They must clarify how MPLS supports virtual private
networks.
• Two models exist, one based on BGP and the other on virtual
routers.
• Protocols like RSVP, OSPF, and IS-IS must be extended
to realize the full benefit of MPLS.
51
MPLS Evolution
• Major efforts are underway to adapt the control
plane of MPLS (e.g., OSPF, IS-IS, LDP, etc) to direct the
routing of optical switches, not just LSRs (label
switched routers).
• This will allow optical switches, LSRs and regular IP
routers to recognize each other.
• The same routing system can control optical paths in
the DWDM core, LSPs (label switched paths) across
the MPLS backbone and any IP routers at the edge of
the network.
52
MPLS Evolution
• With MPLS, service providers can simplify
their operational procedures, deliver more
versatile IP services and sign meaningful SLAs.
53
Key Internet Developments
RTP, RTCP, RTSP (F.Ch28
Multimedia A/V, VOIP)
• RTP (Real-Time Transport Protocol) for audio,
video, etc.
• RTCP - Real-Time Control Protocol
• RTP & RTCP standardized by ITU H.225
• RTSP - Real-Time Streaming Protocol
• VOIP (SIP, H.323)
54