Brkcol 2616

QoS Strategies and Smart Media
Techniques for Collaboration Deployments
BRKCOL-2616
Stefano Giorcelli Glen Lavers

Technical Leader Technical Marketing Engineer
Abstract
In this session we will discuss the network infrastructure and application requirements to ensure a high-quality experience
for interactive voice and video collaboration sessions. Quality of Service (QoS) tools in the network are a key component to
providing a solid and consistent collaboration experience, but they are only one of the aspects needed for an end-to-end
solution. Recent trends such as mobility, BYOD, B2B/B2C, cloud services and the evolution of branch offices have the
consequence that voice and video collaboration sessions increasingly traverse unmanaged networks such as the Internet.
Therefore, other techniques such as media resilience and rate adaptation are required to preserve the quality of experience
end-to-end.
This 2-hour session is divided in two main sections:
In the Collaboration Media section we will discuss video and voice traffic characteristics and requirements and introduce
“smart” media techniques such as media resilience and rate adaptation, showing how they help mitigate the impact of
network impairments such as packet loss and delay.
In the QoS Architecture and Design section we will describe the three main QoS functions in the context of Collaboration:
• media identification and classification (trust boundaries, DSCP marking)
• queuing and scheduling (CBWFQ, WRED)
• provisioning and resource control (planning, bandwidth allocation)
We will look at how these functions are applied to all types of Collaboration endpoints (hardware, software, mobile) in
different parts of the network (LAN, WLAN and WAN) and provide configuration examples to illustrate the concepts. We will
also take into account the “smart” media techniques introduced in the previous section to introduce new queuing and
provisioning models that simplify operations and maximize resource usage. We will also provide design best practices for
common deployment scenarios.
INTRODUCTION
Evolution of Collaboration Landscape
On-premise
UC Services Call
Fixed, hardware endpoints Control
Managed networks
Central
Site Cloud Services
Mobile, software endpoints

Managed
Unmanaged networks WAN Internet
MPLS DMVPN
VPN
Remote Sites
HW Endpoints
Software
Clients
Managed vs. Unmanaged Networks
Where do your media packets go?
On-premise Call Control

UC Services How do you preserve user
Central
experience when media
Site traverses the Internet?
Cloud Services
B2B
QoS-
capable
B2C
Managed
WAN Internet
MPLS DMVPN
VPN
Remote Sites Home/Mobile Users

Evolution of Collaboration Media Streams
Temporal
layers
Adaptive
video bitrate
Multi-device Simulcast
sessions multistreaming
Multipoint
Bridge Multipoint
Bridge
Collaboration
data
Active cascading
Deployment Challenges
• Voice has been deployed to very large scale onto integrated IP networks
• Video so far has seen smaller-sized deployments:
• Overlay networks or “islands”
• Smaller numbers of endpoints (e.g., room systems)
• Deploying “pervasive” video is challenging

• Multi-media, mobile endpoints
• Varying bandwidth needs
• Hard to plan/provision
• Cloud services, B2B/B2C communications and remote/mobile access make it

even harder to control the user experience
Elements of QoS
The Media-Network Relationship
Network
Management
EF
WAN
AF41
BE
Provision Monitor
Identify Media Classify Schedule Resource Troubleshoot
Control Optimize
Current QoS Approaches
Media Identification Considerations
IT-controlled endpoints
 Single-purpose hardware devices
 Completely managed by IT
Enterprise
Network  Allowed to directly tag their traffic
(trusted by the network)
User-controlled endpoints
 Multi-purpose devices (PC, mobile)
Enterprise
Network
 Partially managed by IT (at best)
 Not capable or not allowed to
Internet directly tag their traffic
(untrusted by the network)
10
Classification and Scheduling Considerations
 Same DSCP for audio and video
streams of a video call
Audio of
voice call EF Policer PQ – During congestion, audio and video
streams are equally impacted
Audio of
Telepresence CS4
Video of
CBWFQ  Different DSCP’s for audio
CS4
WAN Link
Telepresence streams in video calls vs. voice
Audio of
AF41
calls
Desktop video
Video of
CBWFQ – Media stream identification difficult
Desktop video AF41 for multi-media mobile clients
other queues
 Different queues for immersive/
room system video and desktop
video
– Complex provisioning, sub-optimal
bandwidth usage
11
Provisioning and Resource Control Considerations
 Models based only on static provisioning and
admission control face increasing challenges:
– Location of software/mobile clients
– Multiple call control platforms (including cloud)
– Evolving media scenarios (multi-modal sessions)
– Heterogeneous environments, unmanaged networks
 Video streams are bursty and adaptive, audio

streams are not:
– Wide operational bandwidth range
– Provisioning based on ‘peak rate’ leads to unused
bandwidth
– Dynamic rate adaptation techniques are becoming
more mature
– “Building an End-to-End IP Telephony Network”,
Networkers 2002  Evolving end-user expectations
12
Our Strategy
“Smart” Media Techniques QoS Tools Design & Deployment
LTRF1 LTRF1
P1 P3
P2 P4
P5
P1
P5 EF Audio
P2 P4
... ... ... ... Queue
Encoder
?
Decoder
EF
AF42
WAN Link
OOS (P4) ACK LTRF1
AF42 Video
Encoder Decoder
Queue
AF41
R1 FEC
AF41
LTRF 0111010001
1000011001
Repair-P R1 0001100
1001000100
0011001011
1011110
FEC
1110010101
... ... R2 1011010010

1010010
R2
Leverage media resilience and

• Consolidate mechanisms to
rate adaptation to enable
• Use media resilience to
identify Collaboration media pervasive video deployments
reduce impact of packet loss through:
• Apply rate adaptation to • Evolve classification and • simplified provisioning
reduce network congestion scheduling recommendations
• optimized bandwidth
utilization
“Smart” Media Techniques 5
4 Media resilience techniques
Network indicates congestion preserve user experience
by queuing and/or discarding during congestion
media packets
6
Endpoints dynamically
adapt media streams
Load Threshold
2
Network classifies traffic
Receiver
Sender and applies differential
treatment 3
Endpoints provide ongoing
1 media feedback (delay, loss, ...)
Endpoints identify their media Receiver Feedback
streams to the network
Quality of Service
High-Level View Switch Port Types:
1 Data Center Trusted: incoming DSCP’s are honored
CUCM sends
2 DSCP’s and UDP Untrusted: incoming DSCP’s are reset,
HW endpoints mark their CUCM Media Applications
media ports to except for devices identified via CDP
media traffic with DSCP (dynamic trust extension)
endpoints in config Telepresence Unity
Server Connection
Campus Access 6
Media and Edge
applications are trusted by
HW Endpoints
3 the network and mark their

Network recognizes HW media traffic with DSCP
AF41 EF endpoints and trusts DSCP’s
3000-3049
3050-3099
Cloud
Internet Edge
DMZ B2B
WAN
SW Clients
AF41 EF Internet
3000-3049
3050-3099
Enterprise Network Expressway C Expressway E
Mobile Users
4 5
SW clients send audio/video Network does not trust PC’s, mobile devices:
streams using UDP ports in config DSCP’s remarked based on UDP ports
QoS Tools
Media Identification Approach
Integration
• Integration with Cisco’s SDN controller (APIC-EM)
Network
(future)
• SDN controller distributes remarking policy to network
Accuracy • Best accuracy, works across network boundaries
• Requires Cisco network device along media path
& Value SDN
Baseline HW only
• DiffServ Architecture (EF, AF41, AF42, BE)
DSCP Trust • Works only for trusted endpoints/bridges
• DSCP may get reset at network boundaries
• Separate UDP port ranges for video

streams and audio streams
UDP Port Ranges • Deployable by any software client on
any network
• Limited security and accuracy
Deployability (today)
QoS Tools
Evolution of Classification Recommendations
Previous New
Audio stream
Voice phones Voice phones EF VoIP Telephony
Audio stream CS4 Real-Time

Telepresence
(CTS, TX, EX, C, MX, Profile, SX) Video stream Interactive
CS4
Desktop video Telepresence

(99xx, 89xx, DX) (CTS, TX, EX, C, MX, Profile, SX) Audio stream AF41
EF Multimedia
Video stream AF41 Conferencing
Software/mobile Desktop video
(Jabber clients) (99xx, 89xx, DX)
Audio stream EF “Opportunistic”

Software/mobile
(Jabber clients) Multimedia
Video stream AF42 Conferencing
QoS Tools
Evolution of Queuing Recommendations
Previous New
AF41 WRED thresholds
(i.e., drop AF41 last)
EF
Audio of Audio of
voice call EF Policer PQ IP Phone
EF PQ
Audio of Audio of Video EF
CS4
BW Assigned to LLQ Classes

Telepresence (PQ)
(Policer) Audio of Jabber EF
Video of CBWFQ
CS4
WAN Link
Telepresence
Audio of
Desktop video AF41 Video of Video AF41 Video
CBWFQ
CBWFQ
Video of
AF41
other queues
Desktop video
other queues
Video of Jabber AF42

(i.e., drop AF42 first)
Agenda of This Session
• Technology Trends and Current Issues
Introduction • Strategy Overview
• Video Traffic
Collaboration Media • “Smart” Media Techniques
(Media Resilience and Rate Adaptation)
• Identification and Classification

QoS Architecture • Queuing and Scheduling
• Provisioning and Resource Control
• Mobile and Remote Access

Design and • B2B, B2C and Hybrid/Cloud Services
Deployment • Example Enterprise Deployment
19
COLLABORATION
MEDIA
Collaboration Media Agenda
• Video Traffic (i.e., a network administrator’s concerns with interactive media/video)
• Bandwidth
• Packet Loss
• Delay
• “Smart” Media Techniques (i.e., how video can be ‘easier’ on the network)
• Media Resilience
• Rate Adaptation
21
Video Traffic: Requirements and Profiles
AUDIO Bandwidth
 Bandwidth:
– Constant bitrate (smooth)
– Small footprint
– Narrow operational range (1:6)
Time
G.729 (24 kbps)  Loss-sensitive
AAC-LD (160 kbps)
 Delay-sensitive
Operational bandwidth
VIDEO Bandwidth  Bandwidth:

– Variable bitrate (bursty)
– Medium/large footprint
– Wide operational range (1:40)
Time
 Loss-sensitive
240p15 (150 kbps) 1080p60 (6 Mbps)
 Delay-sensitive
Operational bandwidth
Video Traffic 1 I-Frame
“Intra-coded” picture
Video Encoding Basics
 Entire picture encoded
1 as a static image
 No reference to other
frames
2 P-Frame
“Predicted” picture
 Based on a previously
2 encoded frame ( 1 )
 Only the differences
from that frame are
encoded
3 P-Frame
“Predicted” picture
3  Reference for
prediction can be
another P-Frame ( 2 )
23
Video Traffic
Audio vs. Video Packet Distribution
Audio Packets Video Packets
P-Frame I-Frame P-Frame
1400 1400
1000 1000
Bytes
600 600
Audio
Samples
200 200
20 ms Time 33 ms
Video Traffic
Bandwidth Usage: High-definition Video Call
3500
I-Frames
3000
2500
Bandwidth (kbps)
2000
1500
1000
500
0
Time (s)
HD video call, 720p30 @ 1920 kbps (1792 kbps video + 128 kbps audio)
Video bandwidth shown (including L3 overhead)
25
Video Traffic
Causes of Packet Loss
#1 Network Congestion Faulty or Misconfigured
• Buffer overflow Network Equipment
• Bandwidth bottlenecks
Si
Si
L1/L2 Transport Issues

• WAN transmission errors
Network Failure, • Wireless interference/collisions
Reroute
26
Video Traffic
Impact of Packet Loss on a Video Stream Frozen video
I1 Artifacts I1
Video
Pulsing
P1 P3 P1
P2 P4 P5 P2 P4 P5
Encoder
... ... ... ... Decoder
?
P5 I1 I1 I1 P4 P3 P2 P1
Out of Sync (OOS)
 Loss of a P-frame triggers request for a new I-frame

– Encoding and transmitting large I-frame takes time
– If any of the I-frame packets get lost, the process needs to restart
– I-frame creates burst that risks exacerbating network congestion (more packet loss!)
 Flickering/pulsing of video when new I-frame arrives
– Video freeze or artifacts when multiple packets are lost
27
Video Traffic
Delay Requirements—Meeting the Human Need
Users
very
satisfied
Si Si
Users
satisfied One-way
Network
Some Latency WAN
users
dissatisfied
Many Si Si
users
dissatisfied
Nearly all
users
dissatisfied
Video Traffic
Delay in Home/Consumer Access Scenarios: “Bufferbloat”
 Delay seen by 1 audio stream

sharing a queue with 1 TCP flow
 100% link utilization
Queue
 Queuing delay: ~100-500ms ! Never
Emptied
http://en.wikipedia.org/wiki/Bufferbloat
 Delay seen by 1 audio stream

sharing a queue with 10 TCP flows
Queue
 100% link utilization Never
Emptied
 Queuing delay: ~400-500ms !
29 29
Collaboration Media Agenda
• Video Traffic (i.e., a network administrator’s concerns with interactive media/video)
• Bandwidth
• Packet Loss
• Delay
• “Smart” Media Techniques (i.e., how video can be ‘easier’ on the network)
• Media Resilience
• Rate Adaptation
30
“Smart” Media Techniques
Goals and Solutions
Goals Mechanisms
Make network congestion Media Resilience
less likely to occur Encoder Pacing
LTRF with Repair

Recover more efficiently
from packet loss FEC
Optimize use of available

Rate Adaptation
network resources
31
Media Resilience
Encoder Pacing
1400
P-Frame I-Frame P-Frame P-Frame
 Each frame must be

1000 packetized onto the wire
Bytes in 33 ms
600
 Endpoint packet
scheduler disperses
200
packets as evenly as
33 ms Time possible
1400
P-Frame I-Frame P-Frame
 Large I-frames may
1000 need to be “spread” over
Bytes 2 or 3 frame intervals
600 Encoder may then ‘skip’ 1-2
frames to stay within bitrate
budget
200
33 ms Time
Media Resilience
Long Term Reference Frame (LTRF) with Repair
Long-Term Reference Frame
LTRF1 (not actually sent on the wire) LTRF1
P5
Repair P-Frame P5
P1 P3 Built from last sync’ed LTRF P1
P2 P4 P2 P4
Encoder
... ... ... ... Decoder
?
P5 P4 P3 P2 P1
OOSLTRF1
ACK (P4)
 Keep encoder and decoder in sync with active feedback messages

– Encoder instructs decoder to store raw frames at specific sync points as Long-Term Reference
Frames (part of H.264 standard)
– Decoder uses “back channel” (i.e. RTCP) to acknowledge LTRF’s
 When a frame is lost, encoder creates a “Repair” P-frame based on the last
synchronised LTRF instead of generating a new I-frame
Media Resilience
Forward Error Correction (FEC)
Encoder Decoder
LTRF Binary XOR R1 FEC

011101000
110000110
Repair-P R1 010001100 100100010
000110010
FEC Binary
111011110
111001010 XOR
... ... R2 110110100
101010010
R2
 Allows decoder to recover from limited amount of packet loss without losing
synchronization
 Can be applied at different levels (x FEC packets every N data packets) to protect
“important” frames in lossy environments
 Correction code can be basic (binary XOR) or more advanced (Reed-Solomon)
 Trade-off is bandwidth increase—best suited for non-bursty loss
34
Rate Adaptation
Key Idea
Video Packet
Bitrate Loss
SLOW
DOWN
Sender t1 t2 t1 t2 Receiver
RR 2
1
3
RTCP
 Receiver observes delay and packet loss over periods of time and signals back
using RTCP Receiver Reports (RR)
 Reports cause the sender to adjust bitrate so as to adapt to network conditions
(downspeeding, upspeeding)
 Two approaches possible:

– Sender-initiated adjustment based on RTCP Receiver Reports
– Receiver-initiated adjustment via call signaling (H.323 flow control, TMBRR,
SIP Re-invite) or explicit request in RTCP message
Rate Adaptation
Generic Loss-Based Algorithm Start
Choose starting bitrate

based on heuristics
Optional startup phase
(faster adaptation)
Observe
Receiver Reports
no Packet loss > Ld% Packet loss < Li% no

for Td seconds? for Ti seconds ?
yes yes
Decrease bitrate by Increase bitrate

Bd% (up to Bmin) by Bi% (up to Bmax)
36
Rate Adaptation
Research and Standards Work
Network-Assisted Dynamic Adaptation (NADA)
Simulation Results
• No existing standards for interactive
video adaptation
• In 2013 IETF created RTP Media
Congestion Avoidance Techniques
(RMCAT) working group, with focus on
RTCWeb:
– Several adaptation algorithms being
studied
– Cisco actively involved
– Simulation results for Cisco’s NADA
algorithm prove it is possible to achieve
‘weighted’ fairness across sessions and
co-exist with TCP
Support in Cisco Collaboration Devices
Encoder Rate
Endpoint / Bridge FEC LTRF Repair
Pacing Adaptation
89xx, 99xx future future --
DX future future
WebEx future
TX/IX future
Jabber
C/EX/MX/SX/Profile
TS (3.1) (3.1)
MCU (4.5) (4.5)
ClearPath
38
Key Takeaways
• Burstiness of traffic and mobility of the endpoints make deterministic
provisioning for interactive video difficult for network administrators
• Media resilience mechanisms help mitigate impact of video traffic on the

network and impact of network impairments on video
• Dynamic rate adaptation creates an opportunity for more flexible provisioning

models for interactive video in Enterprise networks
• Media resilience and rate adaptation also help preserve user experience when
video traffic traverses the Internet or non-QoS-enabled networks
39
QOS ARCHITECTURE:
IDENTIFICATION & CLASSIFICATION

QUEUING & SCHEDULING
PROVISIONING & RESOURCE CONTROL
QoS Architecture
Network
Building an End-to-End Strategy Management
EF
WAN
AF41
BE
Provision Monitor
Identify Classify Schedule Resource Troubleshoot
Control Optimize
• Trusted devices • Map identified • Low Latency • Bandwidth • Prime Collaboration

traffic to its correct Queuing (LLQ) Provisioning Monitoring Assurance
• Untrusted devices
QoS marking • Priority Queuing and Diagnostics
• Admission Control
(PQ)
• QoS marking: • Monitors Voice, Video,
• Class Based
COS, DSCP Inventory and Fault
Weighted Fair
Queue (CBWFQ) Management
• Weighted Random BRKCOL-2320 - Cisco

Early Detection Prime Collaboration - Plan,
(WRED) Design, Implement for
Success
Identification and Classification Agenda
• Trusted Devices Identify Classify Schedule Provision Monitor
• Untrusted Devices
42
QoS Classes
Cisco’s RFC 4594-Based Recommendations
Per-Hop
Application Class Queuing & Dropping Application Examples
Behavior
VoIP Telephony EF Priority Queue (PQ) Cisco IP Telephony (voice)
Broadcast Video CS5 (Optional) PQ Cisco IP Video Surveillance / Cisco Enterprise TV
Realtime Interactive CS4 (Optional) PQ Cisco TelePresence (Legacy CTS-series)
Multimedia Conferencing AF4 BW Queue + DSCP WRED Cisco Telepresence, Jabber, UC Video, WebEx
Multimedia Streaming AF3 BW Queue + DSCP WRED Cisco Digital Media System (VoDs)
Network Control CS6 BW Queue EIGRP, OSPF, BGP, HSRP, IKE
Signaling CS3 BW Queue SIP, H.323, SCCP
Ops / Admin / Mgmt (OAM) CS2 BW Queue SNMP, SSH, Syslog
Transactional Data AF2 BW Queue + DSCP WRED ERP Apps, CRM Apps, Database Apps
Bulk Data AF1 BW Queue + DSCP WRED E-mail, FTP, Backup Apps, Content Distribution
Best Effort DF Default Queue + RED Default Class
Scavenger CS1 Min BW Queue (Deferential) YouTube, iTunes, BitTorent, Xbox Live
Campus QoS Design Considerations
Trust Boundaries
Trust Boundary
Access-Edge Switches
Conditional [MLS] mls qos trust device cisco-phone | cts
Trust
[MQC] trust device cisco-phone | cts
Examples: IP Phone + PC
[MLS] mls qos trust dscp

[MLS] mls qos trust cos
Full Trust
[MQC] default no config necessary
Examples: Collab Server, MCU, TS
Trust Boundary
[MLS] no mls qos trust
No Trust [MQC] no qos trust dscp

Examples: PC, Mac, Handheld
44
Trust Boundaries
• Conditional Trust: Dynamic trust state, where the applied trust state for a port can dynamically
toggle, depending on a successful endpoint identification exchange and the configured endpoint
trust policy. [mls] qos trust device cisco-phone
• Trusted
• Trust CoS: A port in this trust state accepts the 802.1p CoS marking of a 802.1Q tagged packet and
use this value—in conjunction with the CoS-to-DSCP mapping table—to calculate an internal DSCP
value for the packet. [mls] qos trust cos
• Trust DSCP: A port in this trust state accepts the DSCP marking of a packet and sets the internal
DSCP value to match. [mls] qos trust dscp
• Untrusted: A port in this trust state disregards any and all Layer 2 or Layer 3 markings that a
packet may have and generates an internal DSCP value of 0 no [mls] qos trust.
• Classification and Queuing: Layer 3 remarking and queuing based on criteria such as QoS
marking, L4 ports, etc…
45
Trust States and Operations
CoS = 5 CoS = 0
Untrusted Internal
DSCP = 46 no [mls] qos trust DSCP = 0 DSCP = 0
CoS-to-DSCP Mapping Table

CoS = 5
Trust CoS CoS 0  0 CoS 4  32
DSCP = 46 [mls] qos trust cos CoS 1  8 CoS 5  40
CoS 2  16 CoS 6  48
CoS 3  24 CoS 7  56
mls qos map cos-dscp 0 8 16 24 32 40 48 56
Internal CoS = 5
DSCP = 40
DSCP = 40
CoS = 5 CoS = 5
Trust DSCP Internal
DSCP = 46 [mls] qos trust dscp DSCP = 46 DSCP = 46
46
Trust Boundaries
Trust Boundary
Access-Edge Switches WAN-Edge Router
Conditional
Trust Classification,
Examples: IP Phone + PC
Marking and
Queuing
Full Trust IP WAN
QoS Policy
Examples: Collab Server, MCU, TS
policy-map EGRESS-QUEUING
class VOICE
Trust Boundary
priority percent 10 class-map match-all VOICE

class VIDEO match dscp ef
bandwidth percent 30 class-map match-any VIDEO
No Trust class SIGNALING match dscp af41
bandwidth percent 2 class-map match-all SIGNALING
Examples: PC, Mac, Handheld match dscp cs3
47
Egress Classification and Queuing
Trust EF
WAN
AF41
CS3
! This section applies the policy-map to the Interface

Router (config-if)# service-policy output EGRESS-QUEUING
1 ! Attaches service policy to interface
! This section configures the bandwidth for all collab

traffic
class VOICE 2 ! This section applies the policy-map
! Provisions 10% LLQ to VOICE class match dscp ef
class VIDEO class-map match-any VIDEO
bandwidth percent 30 match dscp af41
! Provisions 30% CBWFQ to VIDEO class match dscp af42
class SIGNALING class-map match-all SIGNALING
bandwidth percent 2 match dscp cs3
! Provisions 2% CBWFQ to SIGNALING class
…
3
48
Trust and Enforcement
Devices and Places in the Network
Conditionally-Trusted Endpoints
Access Distribution Core WAN Edge
Immersive Endpoints
CDP Support required

IOS GW
IP Phones, DX, TC, TX, CTS Mac/PC
Trusted Endpoints / Devices
CME
WiFi AP
PC/GPO MAC Handheld
CUBE
Untrusted Endpoints / Devices
Trusted Devices WiFi

Controller
PC MAC Handheld
UCM IM&P VCS / MCU/TS 49

UCx Expressway
Quality of Service
CDP to establish QoS trust boundary – “Conditional Trust”
Boundary
Trust
1. CDP exchange, trust boundary extended to endpoint
2. Phone sets DSCP (audio EF, video AF41, signaling CS3)
3. Phone re-writes DSCP of Computer port traffic to BE
Cisco Catalyst Series
4. Switch trusts DSCP from endpoint and maps CDP: Access Switches
DSCP  DSCP Mac/PC
IP Phones, DX Series
“TC” Series
CTS, TX, IX Series
Data VLAN = 10 Phone VLAN = 110
Mac/PC 802.1Q/p
1 Commands:
CDP Negotiation
[MLS] mls qos trust device cisco-phone | cts
2 4 [MQC] trust device cisco-phone | cts
Voice = EF | Video = AF41 | DSCP EF (46)
Sig = CS3 For multilayer switching (MLS) the mls qos
3
DSCP AF41 (34)
“trust cos” or “trust dscp” command is required
DATA = EF DATA = 0 DSCP CS3 (24) to determine the type of trust once extended.
DSCP BE (0)
New
(CUCM 11.0)
Custom QoS Settings For SIP Devices
SIP Profile Settings
• New Separation of Audio and Video UDP Port Ranges
• New DSCP parameters
Benefits
• Granular configuration of SIP Endpoint groupings for both Trusted (DSCP
honored) and Untrusted (UDP Port Remarking) Endpoints.
New
(CUCM 11.0)
Custom QoS settings for SIP Devices
Clusterwide Parameters (System – QoS)
DSCP for Audio Calls
DSCP for Video Calls TelePresence
DSCP for Audio Portion of Video Calls Endpoints
DSCP for TelePresence Calls
DSCP for Audio Portion of TelePresence Calls
… TelePresence Applicable DSCP settings:

DSCP for TelePresence Calls
QoS Registration DSCP for Audio Portion of TelePresence Calls
Service SIP Profile Device
Parameters Config File
UC Video
Unified
Endpoints
CM
UC Video Applicable DSCP settings:

DSCP for Video Calls
DSCP for Audio Portion of Video Calls
New in 11.0
TelePresence Endpoints in Unified CM
• TelePresence endpoints are identified as immersive
video endpoints
• Fixed setting (Not Configurable)
• Check Devices for Capability:
• Cisco Unified Reporting Tool > “Immersive Video
Support for TelePresence Devices”
Custom QoS Settings For SIP Devices New
(CUCM 11.0)
SIP Profile (Defaults Modified for Example)

New
Custom QoS settings for SIP Devices (CUCM 11.0)
SIP Profile 2
SIP Profile 1
TelePresence Desktop Jabber

Endpoints Video Clients
Endpoints
Prioritized Video: AF41 “Opportunistic” Video: AF42

DSCP for basic call flows1
TelePresence UC Video Call Flow
Endpoint Endpoint
DSCP for Audio Calls X X Voice-Only
DSCP for Video Calls X Video – Audio and Video stream of a video call
unless the endpoint supports “DSCP for Audio
Portion of Video Calls” parameter
DSCP for Audio Portion of X Audio stream of a video call – Only applicable to
Video Calls2 endpoints which support the parameter
DSCP for TelePresence Calls X Immersive Video- Audio and Video of an immersive
video call unless the endpoint supports “DSCP for
Audio Portion of TelePresence Calls” parameter.
DSCP for Audio Portion of X Audio stream of a video call – Only applicable to
TelePresence Calls2 endpoints which support the parameter
TelePresence UC Video Call Flow
Endpoint Endpoint
1 The DSCP settings for Multi-Level Priority and Preemption (MLPP) are not discussed here. Please
refer to Unified CM System Guide for more information about MLPP and QoS settings.
2 The DSCP for Audio Portion of Video or TelePresence is not supported on all video endpoints at this
writing. See table below for support of these parameters based on the type of endpoint.
Flexible DSCP Endpoint Support
DSCP for Audio Portion of DSCP for Audio Portion New Separate UDP Port
Video Endpoint
Video Call of TelePresence Call Ranges
8800 Series No NA No
8900/9900 Series No NA No
Jabber Clients (All) Yes NA Yes
DX Series Yes Yes1 No
TX Series NA No No
IX Series NA No No
CE 8.x Software Series (SX
NA Yes No
Series, MX G2, MX700/800)
TC 7.1.4 Software Series (C
Series , Profile Series, EX Series NA Yes No
MX G1)
EX Series (TC Software) NA Yes No
1 To enable the DX series to use DSCP for TelePresence Call as well as DSCP for Audio Portion of TelePresence Call when in calls
with TelePresence devices requires the enablement of the Video Promotion feature (see the Unified CM System Guide Video
Telephony chapter for more information on enabling this feature).
Unified CM QoS Classification
Audio Portion of Video/TelePresence calls (*Only supported on specific phone
software releases) Default Values
Unified CM System QoS Values and CAC Pool Associations
Service Parameter Name Media Stream Type DSCP Value PHB Value CAC Pool
DSCP for Audio Calls Audio Only 46 EF Voice
*DSCP forVideo
Audio Portion of
Calls
Audio of Video 34 AF41 Video
DSCP for Video Calls Video of Video 34 AF41 Video
*DSCPTelePresence
for Audio Portion of
Calls
Audio of TP 32 CS4 Video
DSCP for TelePresence Calls Video of TP 32 CS4 Video

Unified CM QoS Classification
Audio Portion of Video/TelePresence calls (*Only supported on specific phone
software releases). Recommended Values for CUCM 11.x deployment
Unified CM System QoS Values and CAC Pool Associations
Service Parameter Name Media Stream Type DSCP Value PHB Value CAC Pool
DSCP for Audio Calls Audio Only 46 EF Voice
*DSCP forVideo
Audio Portion of
Calls
Audio of Video 46 EF Voice
DSCP for Video Calls Video of Video 34 AF41 Video
*DSCPTelePresence
for Audio Portion of
Calls
Audio of TP 46 EF Voice
DSCP for TelePresence Calls Video of TP 34 AF41 Video

Unity Connection QoS
 System settings  Advanced  Telephony
 Sets QoS for audio, video and signaling (SCCP/SIP) separately
 Default = Audio (46 / EF), Video (46 / EF), Signaling (24 / CS3)
 Change Video to 34 / AF41
NOTE: When doing clustering over the WAN for Unity Connection nodes
use the CLI command “utils cuc networking dscp on”, sets intra-cluster
traffic with a DSCP of 18.
60
TelePresence Server QoS (Virtual and HW)
 Network  QoS
 Sets QoS for audio and video separately
 Default = Audio (46 / EF) and Video (34 / AF41)
 Default Recommended
61
MCU QoS Recommended QoS settings on the MCU:
Default QoS settings on the MCU:
 Audio 101110 = EF  Audio 101110 = EF
 Video 100010 = AF41  Video 100010 = AF41
 Streaming 000000 = BE  Streaming 000000
 Signaling 000000 = BE  Signaling 011000 = CS3
 Admin 000000 = BE  Admin 100000 = CS2 MCU 4501
62
TelePresence VCS and Expressway Cisco Expressway Core
 System  Quality of Service (C) and Edge (E)
 Sets QoS for ALL Media and Signaling to a single DSCP
 Default = 0
 See the Design and Deployment section:
(36 / AF42 or 34 / AF41)
63
• Untrusted Devices:
• Mapping UDP/TCP Port Ranges
• SDN (Future)
64
Mapping Identifiable Media and Signaling Streams
1. Identifying media and signaling streams from the client based on Layer 4
(Transport) port ranges (Protocol TCP/UDP and Ports).
a. The protocol port ranges are configured in Unified CM and are passed to the
endpoint during device registration. The endpoint then uses these port ranges
for signaling and media negotiation.
2. Classify the signaling and media streams and remark with a corresponding
DSCP
a. Network Access Control Lists (ACL): Method consists of configuring ACLs to classify
and mark DSCP based on protocol and port ranges
b. Windows Global Policy Objects (GPO): Method consists of configuring GPO’s to
classify and mark DSCP based on protocol, port ranges and application and relies on
DSCP trust to pass through the network
65
Mobile Portfolio: Jabber Clients
QoS Classification and Marking in Jabber Products
• Classification in Windows Vista, 7 and 8
• Global Policy Objects (GPOs) which specifies Protocol, Port and Application as means of
identification of traffic by which to mark QoS
• Classification in Windows 2000 and XP
• Windows 2000 and XP have a different model for allowing the application to mark QoS,
which is called Generic Quality of Service (GQoS). Jabber for Windows has implemented
GQoS allowing the application to inform the OS to mark the desired DSCP. Turning this
function on in Windows 2000 and XP is explained in the following Microsoft Knowledge Base
article: http://support.microsoft.com/kb/248611
• Classification in Mac OSX
• Natively marks DSCP
• Classification in iOS (iPhone and iPad)
• Natively marks DSCP Latest Jabber
• Classification in Android Releases
• Natively marks DSCP
67
Jabber’s use of UDP Port Ranges
Media and Signaling Port Range – Common
Upper Half
1 2 for Video 3
Registration 4000-4999
Lower Half
Config File for Audio
Unified Jabber 3000-3999
CM SEP[MAC_addr].cnf.xml splits media
range in SIP
Media Port Range UDP Jabber Clients
half Signaling
3000-4999
TCP 5060
SIP Signaling Port TCP
5060
1. Client registration, download configuration file

2. Split media port range in half, upper half for video and lower half for
audio
68
Jabber’s use of UDP Port Ranges New
(CUCM 11.0)
Media and Signaling Port Range - Separated
Video
1 2 Range 3
Registration 5000-5999
Audio
Config File Range
Unified Jabber 3000-3999
CM SEP[MAC_addr].cnf.xml uses the
two port SIP
Media Port Range UDP Jabber Clients
ranges for Signaling
3000-3999
media TCP 5060
Video Port Range UDP
5000-5999
SIP Signaling Port TCP
5060
1. Client registration, download configuration file

2. As of version 11.0, CUCM can provide Jabber with two separate port
ranges for audio and video streams
69
Jabber’s use of UDP Port Ranges
3. Configure Network with ACL’s:

Media is identified on UDP port ranges Ingress Services
3xxx and 5xxx and classified and Switch Edge Router
remarked to EF and AF42 respectively.
SIP Signaling is identified on TCP Port
range 5060-5061 to and classified and
remarked to CS3.
Example QoS Policy for Jabber Clients:
• UDP Port Range 3xxx Mark DSCP EF
• UDP Port Range 5xxx Mark DSCP AF42
• TCP Port 5060-5061 Mark DSCP CS3
70
Configuring Jabber Client
Managing Media Ports: SIP Profile
71
Configuring Jabber for Windows Client
Managing the SIP Signaling Port: IP Security Profile
72
Jabber Client Summary / Best Practices
• Use the SIP Profile to configure media port range (default 16384-32766) and
DSCP settings
• Use the SIP Security Profile to configure the signaling port range (default 5060
or 5061 for secure signaling)
• If you have SCCP devices in the same network use a port range outside of
16384-32766 to avoid overlap and incorrect remarking
• Unified CM 9.1 expands SIP media port range to 2048-65535
• Video Enablement:
• Disable video if you do not want device to send or receive video
• Video capable devices ALWAYS divide the port ranges (even if video is disabled)
• Devices that do NOT support video (version dependent) use the entire port range for
audio-only.
73
Ingress Classification
UDP Ports DSCP
Jabber Client Ingress
Policy 1
3xxx EF
ACL ! This section applies the policy-map to the Interface
5xxx AF42 Router (config-if)# service-policy input INGRESS-MARKING
5061 CS3 ! Attaches service policy to interface
! This section configures the ACL’s
access-list 100 permit udp any any range 3000 3999 5 ! This section configures the policy-map to set DSCP
access-list 101 permit udp any any range 5000 5999
for JABBER Voice, Video and SIP Signaling on 2
access-list 102 permit tcp any any range 5060 5061 ingress
policy-map INGRESS-MARKING
class VOICE
4 set dscp ef
! This section configures the classes class PRIORITIZED-VIDEO
set dscp af41
class-map match-all JABBER-VOICE
class JABBER-VIDEO
match access-group 100 set dscp af42 6
class SIP
class-map match-all JABBER-VIDEO 3 set dscp cs3
match access-group 101
class class-default
class-map match-all JABBER-SIP

Egress Queuing
3000 EF
WAN
4000 AF42
5060 CS3

Router (config-if)# service-policy output EGRESS-QUEUING
6 ! Attaches service policy to interface
! This section configures the bandwidth for all collab

traffic
…
8
Jabber For Windows “To Trust or Not to Trust”
Group Policy Objects
Media Port Range
(3000-4999) QoS Group Policy
Audio Call Voice Stream

(Voice Stream)
(EF)
(Ports:3000-3999)
Voice Stream
Video Call
(Voice and Video
Streams) (AF42) Video Stream
(Ports:4000-4999)
SIP Port Range

(5060-5061) QoS Group Policy
SIP (CS3) SIP Signaling

Signaling (Ports:5060-6061)
76
Group Policy Objects (GPO) Policies Illustrating:
Difference from 64-bit and
32-bit Application Paths
Policies Illustrating:
Protocol, ports, IP
Addresses and DSCP
Values
77 77
• Untrusted Devices:
• Mapping UDP/TCP Port Ranges
• SDN (Future)
78
Network Integration – SDN
Dynamic Policy Management for Untrusted Devices (e.g., Jabber Clients)
CUCM
See BRKCRS-3011, “APIC-EM
(Application Policy Infrastructure
Controller - Enterprise Module) –
SDN in the Enterprise”
Tuesday June 9 at 8:00am
Cisco® APIC
Enterprise Module EM
CUCM
See demo at World of Solutions
Application
Dynamic Policy
Management
Jabber Client Jabber Client
Traffic Queuing
QOS ARCHITECTURE:

QoS Architecture
Network
Building an End-to-End Strategy Management
EF
WAN
AF41
BE
Provision Monitor
Control Optimize
• Trusted devices • Map identified • Low Latency

traffic to its correct Queuing (LLQ)
• Untrusted devices:
QoS marking • Priority Queuing
• Map UDP/TCP (PQ)
Port Ranges • QoS marking:
• Class Based
COS, DSCP
Weighted Fair
Queue (CBWFQ)
• Weighted Random
Early Detection
(WRED)
Queuing and Scheduling Agenda
• WAN Queuing Considerations Identify Classify Schedule Provision Monitor
• IOS Queuing (WFQ, CBWFQ, LLQ)
• Congestion Avoidance (WRED)
82
IOS Queuing
Low-Latency Queuing (LLQ)
IOS Interface Buffers
1 Mbps
VoIP policy-map LLQ
Policer
class VOIP
LLQ
priority 1000
…
Packets
In Packets
Out
CBWFQ
Scheduler
Tx-Ring
FQ CBWFQ
Pre-Sorters
See Network Infrastructure chapter of upcoming SRND 11.0

Congestion Avoidance random-detect dscp <dscpvalue> <min-threshold> <max-
threshold> [mark-probability-denominator]
Weighted Random Early Detection (WRED)
Tail Front
of of
Queue Queue
Desktop Video CBWFQ
Fair-
Queue
Direction
Pre- of
Sorter Packet
Flow
AF43 Minimum WRED Threshold:

Begin randomly dropping AF43 Packets
AF42 Minimum WRED Threshold: policy-map BULK-WRED

Begin randomly dropping AF42 Packets class DesktopVideo
bandwidth percent 10
AF41 Minimum WRED Threshold: random-detect dscp-based
Begin randomly dropping AF41 Packets random-detect dscp 34 100 128
random-detect dscp 36 90 128
random-detect dscp 38 75 128
Maximum WRED Thresholds for AF41, AF42 and AF43 are set to the tail of the queue in this example
See Network Infrastructure chapter of upcoming SRND 11.0

Congestion Avoidance
WRED Configuration Example: 10 Mbps WAN Interface
100%
Drop Probability
policy-map WAN-EDGE
class VIDEO
~ ~ bandwidth percent 40
random-detect dscp-based
~
random-detect dscp 34 1200 256 50
mark-probability-denominator
5%
WRED AF42
mark 2%
probability AF41
(100/denom)
30 80 120 256
WRED
min threshold Queue Occupancy (packets)
WRED
max threshold See Network Infrastructure chapter of upcoming SRND 11.0
85
Congestion Avoidance
How to choose WRED Parameters?
WAN Link Speed 622 Mbps 155 Mbps 34-44 Mbps

10 Mbps 5 Mbps
WRED Values (OC12) (OC3) (E3/DS3)
min-threshold 240 180 120 120 60
AF41 max-threshold 512 384 256 256 128
mark-probability-
50 50 50 50 50
denominator
min-threshold 40 30 30 30 15
AF42 max-threshold 180 135 80 80 40
mark-probability-
20 20 20 20 20
denominator
Video queue bandwidth % 43 53 55 55 30
86
WAN Queuing Considerations
Summary
(i.e., drop AF41 last)  Map audio streams of voice and
video calls (EF) to a priority queue
EF
Audio of
IP Phone
EF PQ
 Map video streams of video calls
Audio of Video EF (AF41 and AF42) to a class-based

Audio of Jabber EF queue with WRED:
– AF41: higher drop thresholds
(e.g., 50-100% of queue depth)
AF41
Video of Video Video
CBWFQ
– AF42: lower drop thresholds
(e.g., 15-35% of queue depth)
other queues
Video of Jabber AF42  During congestion, AF42 traffic is
dropped first:
– Packet loss triggers rate adaptation
(i.e., drop AF42 first) – Media resilience limits the impact
WAN Queuing Considerations
Summary
Question:
Why implement a single CBWFQ with WRED instead of multiple separate
CBWFQ’s?
Answer:
Because unused bandwidth from one class of video can be fully used by another
class of video if both classes are in the same CBFWQ.
For example, Jabber video can use the unused bandwidth from room system
video if they are in the same CBWFQ. If room system video were in a separate
CBWFQ, then unused bandwidth from that queue would be equally shared across
ALL other queues!
QOS ARCHITECTURE:

QoS Architecture
Building an End-to-End Strategy Network
Management
EF
WAN
AF41
BE
Provision Monitor
Control Optimize
• Trusted devices • Map identified • Low Latency • Bandwidth

traffic to its correct Queuing (LLQ) Provisioning
• Untrusted devices:
QoS marking • Priority Queuing
• Map UDP/TCP • Admission Control
(PQ)
Port Ranges • QoS marking:
• Class Based
COS, DSCP
Weighted Fair
Queue (CBWFQ)
• Weighted Random
Early Detection
(WRED)
Provisioning and Resource Control Agenda
Identify Classify Schedule Provision Monitor
• Bandwidth Provisioning
• Admission Control: Enhanced Location CAC

A word about Regions…
• CUCM Locations work in conjunction with Regions to define the characteristics of a call flow
• Regions define the type of compression or bit rate (8 kbps or G.729, 64 kbps or G.722/G.711,
and so forth) that is used between devices
• Location links define the amount of available bandwidth for the path between devices.
• You assign each device in the system to both a region (by means of a device pool) and a
location (by means of a device pool or by direct configuration on the device itself).
• Building region matrix to manage max video bit rate (video resolution) for groups of devices.
• Group devices into max video bit rate categories
• The smaller the number of groups the easier to calculate bandwidth requirements
• Consider the default region settings to simplify the matrices!
92
Region Matrices for Max Video Bit Rate
Example: Matrix with 4 groups
• Re-use regions configured for audio-only IP phones

• Audio codec configuration is shared, so if video calls need to use different audio codecs you
need to configure new regions
• Consider the default region settings to simplify the matrix
• Per-site regions may not be needed if a single audio codec is used for both intra-
region and inter-region calls
Endpoint Legacy (Small Room System +
Jabber Immersive + MCU
Groupings Screen) Smart Desktop
Legacy (Small Screen) 800 800 800 800
Jabber 800 1500 1500 1500
Room System + Smart

800 1500 2500 2500
Desktop
Immersive + MCU 800 1500 2500 12000

93
Region Matrices for Max Video Bit Rate
Absolute Maximums per type
• Create a simple table showing examples of region video bandwidth settings for
different types of endpoints:
– E.g. Jabber 768k, single-screen room systems (MX, Profiles) 1500 kbps, etc…
Endpoint type absolute maximums
Jabber 1200
Single-screen video 1920 kbps

room systems (MX,
Profile)
Smart desktop 1200 kbps
endpoints (DX, EX)
Video IP phones 384
3-screen immersive 6 Mpbs

telepresence
MCU/TS 6 Mbps
94
Tested average bandwidth utilization
TelePresence endpoints, Jabber and DX series
Resolution Endpoint MX200 SX20 EX90 TX9000
Detail Lowest Highest Lowest Highest Lowest Highest Lowest Highest
720p30 (1280x720)
736 kbps 1.2 Mbps 812 kbps 1.2 Mbps 812 kbps 1.2 Mbps 3.1 Mbps 6.4 Mbps
1080p30 (1920x1080)
2.6 Mbps 5.7 Mbps 2.6 Mbps 6.2 Mbps 2.5 Mbps 6.1 Mbps 8.8 Mbps 11.9 Mbps
720p60 (60fps)
- 2.3 Mbps - 2.3 Mbps - 2.4 Mbps - -
DX series video
Jabber bandwidth Resolution
Resolution bandwidth range
(with G.711 audio)
240p30 (432x240) 150-299 kbps
w144p30 (256x144) 156 kbps
360p30 (640x360) 300-599 kbps
w288p30 (512x288) 320 kbps
480p30 (848x480) 600-799 kbps
w448p30 (768x448) 570 kbps
576p30 (1024x576) 800 kbps-1.29 Mbps
w576p30 (1024x576) 890 kbps
720p30 (1280x720) 1.3-1.99 Mbps
720p30 (1280x720) 1.3 Mbps
1080p30 (1920x1080) 2-4 Mbps
http://www.cisco.com/en/US/partner/prod/collateral/ps7060/tested_bandwidth_whitepaperx.pdf
http://www.cisco.com/en/US/partner/docs/voice_ip_comm/jabber/Windows/9_6/InstallConfig/JABW_BK_CDFE9752_00_installation-and-configuration_chapter_01.html#JABW_RF_Q07E13E6_00
http://www.cisco.com/en/US/partner/docs/voice_ip_comm/cdce/dx600/admin/10_1_1/english/DX60_BK_CFB047D4_00_cisco-dx600-administration-guide-10_1_1_chapter_01000.html#DX60_RF_V3296974_00
95
Provisioning and Resource Control Agenda
Identify Classify Schedule Provision Monitor
• Bandwidth Provisioning
• Admission Control: Enhanced Locations CAC

• Network Modeling
• Locations Bandwidth Manager (LBM)
• Inter-Cluster E-LCAC with LBM
Control Plane
NO
E-LCAC
YES
?
Call Admission Control
QoS
Scheduling + Queuing
IP WAN
Data Plane
LCAC Limitations E-LCAC Solutions
Limited WAN Topology Support: Network Modeling:
• Hub and Spoke WAN topology support • Convert UCM locations to a model capable
• Large gap between RSVP and Locations of supporting real network topologies
CAC
Multi Cluster Support: Inter-Cluster (Inter-cluster) CAC:

• Multiple Clusters that managed endpoints in • Implement a bandwidth-accounting
same branch sites could only inefficiently scheme that works between multiple
subdivide inter-branch bandwidth to avoid Unified CM clusters and dynamically
quality degradation (Ships in the night CAC)
learns the topology from one another
TelePresence: Immersive Bandwidth Allocations:

• Did not Support CAC (overlay design) • Implement an immersive BW pool in locations
• TelePresence and UC or 3rd party video on a CAC
single cluster • Provide better CAC interop support between
• Limited CAC support for TelePresence video TelePresence Video and Desktop Video
interoperability (P2P calls without an MCU)
Network Modeling - Concepts A
Location Deduct BW!
• Administrator builds a Network Model using locations and
links
• A Location represents a LAN. It could contain endpoints B C
or simply serve as a transit location between links for WAN
network modeling
• Links interconnect locations and are used to define Deduct BW!
bandwidth available between locations. Links logically Link
represent the WAN link
• Weights are used on links to provide a “cost” to the
“effective path”. Weights are pertinent only when there is D E
more than 1 path between any 2 locations
Deduct BW!
• UCM calculates shortest paths (least cost) from all
locations to all locations and builds the effective paths
• The Effective paths are the paths with the “least F
cumulative weight”
• UCM tracks bandwidth across any link that the network
model indicates from originating Location to terminating
location. Effective Path
99
Network Modeling – Locations and Links
Location A BETWEEN Location B
Link A < > B BW Allocated
BW Allocated
Audio Unlimited BW Allocated Audio Unlimited
Video 100MB Audio 1500k Video Unlimited
Immersive 250MB Video 3000k Immersive Unlimited TelePresence

Immersive 5000k
IN/OUT Links Provide Bandwidth IN/OUT

Accounting Between
Locations
And Interconnect Locations
WITHIN WITHIN
Locations Provide Bandwidth TelePresence and UC Video

Accounting WITHIN the Linking Locations Endpoints Can Reside in the
Location as well as IN or OUT Same Location*
of the Location
EUG PDX SEA BLD 100
Intra-Location Bandwidth Allocation – TelePresence Immersive
• Links Interconnect Locations to Build the Topology. Bandwidth Values and Weight are Assigned to Links
• Intra-location Bandwidth Limits are Assigned to a Location to CAC ALL calls made TO/FROM/WITHIN the Location.
Intra-location Bandwidth Values are Unlimited by Default.
PDX As Viewed From

The Perspective of
The PDX Location.
Serviceability
Provides More
BLD EUG SEA YVR
Tools for Topo
Visibility
101
The Location Admin Page Has Been Updated To Configure Location Links
• By default when a new location is created a link to Hub. None will be added with
unlimited audio bandwidth, 384 kb of both video and immersive bandwidth
• RECOMMENDATION: DELETE the link when it’s not needed
Network Modeling in Locations CAC
Links, Weights and “Effective Path”
 Weight provides the ability to force a specific path choice
Effective Path when multiple paths between 2 locations are available
San Jose
 When Multiple Paths are configured yet only 1 will be
selected based on “Weight”. This path is the “Effective
Path”
Link
Weight  Weight is used to determine path cost, lowest weight path
from source to destination is selected
 Weight is static and does not change with regards to the

50 “effective path” from one location to another
30
EFFECTIVE PATH
30 Path 1:
San Jose > Seattle (Weight = 50 = 50)
Path 2:
San Jose > Boulder > Seattle (Weight = 30 + 30 = 60 )
Boulder Seattle
Network Modeling in Locations CAC
Links, Weights and “Effective Path”
• The Locations Bandwidth Manager (LBM) service computes the effective path
from source location to destination location:
 Sum weight of links across each possible path from source to destination
 The least cost value of the path’s weight determines the “Effective Path”
 A tie break of equally weighted paths is determined by LBM based on location name
 Once the effective path is determined, all subsequent calls that have the same source and destination
locations will use the same “Effective Path”
Serviceability > Tools > Locations > Effective Path: Provides the Ability to Ascertain
the “Effective Paths” Configured in the Topology
New
New Enhanced Locations CAC Feature (CUCM 11.0)
• Call Admission Control Enhancement for Audio Portion of Video Calls:
When enabled CAC for video calls deducts the audio portion of the video call
from the voice pool and the video portion of the video call from the video pool (or
immersive pool if it’s enabled).
• Custom QoS settings for SIP Devices ensure that the audio of video and
immersive calls are set to the correct DSCP (EF) to align with the CAC
bandwidth deduction.
Network Modeling
Key Takeaways
• Enhanced Locations CAC is a Static Model-Based CAC Mechanism
• E-LCAC is a Model of the “Routed Network” Attempting to Represent How The
WAN Network Topology Routes Media
• Network Modeling is NOT Dynamic like RSVP
• The Model Needs to be Updated When the Network Topology Changes
• E-LAC is Call-Based (No Asymmetric or Unidirectional Bandwidth Deductions)
• Intra-location bandwidth assignment and deduction. The default is set to
unlimited.
UCM
Location Bandwidth Manager (LBM)

Service Overview LBM LBM
Co-Located StandAlone
• LBM is a Unified CM Feature Service
• LBM Service is Enabled by Default When Upgraded from a Pre-9.0 Installation
• For Fresh Installs The LBM Service Needs to be Manually Activated (like CCM service)
• LBM Can Run on Any UCM Subscriber or Standalone
• For E-LCAC to Function LBM Must Be Enabled
• Functions of LBM:
• Location Path Assembly and Calculation
• Servicing Bandwidth Requests from Unified CM Call Control (XML/TCP)
• Replication of Bandwidth Information to Other LBMs Within and Between Clusters (Inter-Cluster
Locations CAC)
• Provides Configured and Dynamic information to Serviceability
• Updates Location RTMT counters
LBM
XML/TCP
UCM
Location Bandwidth Manager
Communication and LBM Replication
• LBM Services Within a Cluster Are Always Fully 4 node Cluster

Meshed and Replicate Bandwidth Allocations
UCM
• The CallManager Service Communicates with The BW Req/Res

LOCAL LBM Service (Default)
LBM UCM
• Recommendation: UCM
• Run LBM on same node as Cisco BW Req/Res

BW Req/Res
CallManager Service
LBM
LBM
LBM
BW Request / Response (XML/TCP) LBM Full Mesh
LBM Replication (XML/TCP) Replication
BW Req/Res
UCM
Inter-Cluster Enhanced Locations CAC
• Extends Enhanced Locations CAC Network Modeling Across Multiple Clusters
• Each Cluster Manages Its Own Topology
• Each Cluster Then Propagates Its Topology to Other Clusters Configured In the LBM
Inter-Cluster Replication Network
• Each Cluster Then Creates a Global Topology (“Assembled Topology”) Piecing
Together Each Clusters Replicated Topology
Assembled / Global Topology
Hub_none
LBM
Replication
Cluster 1 Cluster 2
Hub_none Hub_none
Loc_11 Loc_12
Loc_21 Loc_22 Loc_24
Loc_25
Loc_11 Loc_12
Loc_21 Loc_22 Loc_24
Loc_25
Cluster 1 Cluster 2
Configured Topo Configured Topo
LBM Network – Hubs, Spokes and Hub Bootstrap
Centralized Hub Bootstrap server for the LBM replication network
• SME Servers As the Centralized SME Cluster LBM Hub Group
Hub
Hub Bootstrap Hub Servers LBM Hub Servers
Servers
LBM Bootstrap Hub Servers (Bootstrap)
• With Other Remote Hubs as SME_1 SME_2 SME_1
Backup Hub Bootstrap Servers SME_2
Leaf Cluster 1 Member3
Leaf Cluster 2
HUB HUB LBM Hub Group
LBM Hub Group
Hub Hub
LBM Hub LBM Hub
Servers Servers
Servers Servers
(Bootstrap) (Bootstrap)
SME_1 SME_1
UCM_3 UCM_1 UCM_ SME_2
UCM_C
SME_2
A
NONE
UCM_1
NONE
UCM_A
HUB HUB LBM

LBM
HUB HUB LBM

LBM
UCM_4 UCM_2 UCM_ UCM_D

B
Inter-Cluster LBM Replication Network
Leaf Cluster 1 Replication Network Hubs Only Leaf Cluster 2 Replication Network
110
Audio and Video Admission Control
Considerations
No Admission Control
• Over-provision queues
• Rely on Video Rate Adaptation and Media Resiliency Capabilities
• Audio is much easier to over-provision in pervasive video deployments
• QoS is critical and rate adaption is highly beneficial for both managed/unmanaged
networks
• Benefits: Simplicity
Admission Control
• Strict provisioning (Mapping CAC to Queuing)
• Mobility? Device Mobility feature (Adds OPEX)
• Benefits:
• Manage lower bandwidth links, use Automated Alternate Routing (AAR) for PSTN redirect
• Ensure quality audio during the busy hour by avoiding oversubscription and packet loss
• Safe when over-provisioning is not an option
111
DESIGN AND
DEPLOYMENT
Design Considersations Agenda
• B2B, C2B and Cloud Interactions
• Example Enterprise Deployment
113
Mobile and Remote Access
Media Path Summary
Media Traversal
• “C” calls “A” on-premise
QoS-enabled • Expressway solution provides firewall
Enterprise Network B
traversal for media
Inside firewall DMZ Outside firewall • Expressway C de-multiplexes media

and forwards toward “A”
(Intranet)
Media Relay
Collaboration
Internet
Services C • “C” calls “B” off-premise
UCM Expressway Expressway • Media is relayed via Expressway C

C E
Optimized Media (future ICE support)
D • “B” calls “D” off-premise

SIGNALING
• Both “B” and “D” are ICE-enabled
A MEDIA
• STUN binding success
• Media flows are optimized between
endpoints
Mobile and Remote Access
QoS Best Practices
• All media coming from remote endpoints is relayed via Expressway C*
• Expressway C can mark traffic with a single DSCP
• no distinction between audio and video streams
• Recommendations:
• Configure Expressway C to mark traffic with DSCP AF42 (36)
• Configure DSCP trust on access switch port connected to Expressway C
• (optional) if CAC is deployed,
configure Device Mobility using
Expressway C’s IP addresses
and assign endpoints to an
“Internet” location
*: Same considerations apply if using VCS C/E to

provide the Mobile and Remote Access feature
115
Design Considerations Agenda
116
B2B, C2B and Cloud Interactions
B2B and C2B Considerations
 Voice/video traffic is relayed

Jabber Guest CUCM Company XYZ
into the Enterprise network
Server via VCS/Expressway C/E
Ad-hoc B2B
 DSCP marking options:
On-premise
– AF42 (36) for all traffic from
VCS-C VCS-E
Conferencing
(MCU/TS)
VCS/Expressway C
Internet – AF41 (34) for all traffic from
VCS/Expressway C
– Best Effort (0) for all traffic
from VCS/Expressway C
Enterprise
 If using CAC for B2B/C2B calls,
Network
C2B remember to place CUCM trunks to
(Jabber Guest) VCS-C/Expwy C in the right location
WebEx Considerations: CMR Hybrid and CMR Cloud
 Voice/video traffic is relayed

CUCM
WebEx into the Enterprise network
CMR Hybrid
via VCS/Expressway C/E
 DSCP marking options:
On-premise
Conferencing
(TS) Expwy-C Expwy-E
– AF42 (36) for all traffic from
Internet VCS/Expressway C
Company A – AF41 (34) for all traffic from
VCS/Expressway C
CUCM
WebEx  For CMR cloud, AF41
CMR Cloud
recommended for
Expwy-C Expwy-E environments where CAC is
configured
Company B Internet
WebEx Considerations: VoIP and Video in WebEx Meetings Client
Enterprise Network
WebEx • Video in WebEx Meetings

Client IP WAN client is rate-adaptive
Internet
Branch office HQ
• To provide better-than-
Protocol Port Number WebEx Access Type best-effort QoS when
TCP 80 WebEx Client Access traversing Enterprise
TCP 443 WebEx Client Access - Secure Traffic (SSL Sites) WAN:
TCP/UDP 1270 WebEx Client Access (Non SSL Sites)
• classify traffic with an ACL
based on UDP ports 9000
TCP/UDP 53 WebEx Domain Name System (DNS)
and 9001
TCP/UDP 5101 WebEx MMP
• remark it with DSCP AF42
TCP 8554 WebEx Audio Streaming Client Access
UDP 7500-7501 WebEx Audio Streaming
UDP 9000-9001 WebEx VoIP/Video

Design Considerations Agenda
120
Example Enterprise Deployment
CUCM
Multipoint
Video (TS) Central
500 users:
Site
500 Jabber clients
300 IP phones Expressway B2B
30 video endpoints
(room + desktop) Cloud
1 immersive telepresence Services
2 Telepresence Servers
Mobile
Internet Users
MPLS VPN
50 users: 15 users: 5 users:

50 Jabber clients 15 Jabber clients 5 Jabber clients
30 IP phones 10 IP phones 3 IP phones
6 video endpoints 2 video endpoints 1 video endpoint
(room + desktop) (room)
Large Branch Small Branch Micro Branch
Identification and Classification
• Jabber classification based on
UDP Ports
UDP port ranges and ACL’s:
DSCP
• Audio streams of all calls (voice-
only and video) marked EF
3xxx EF • Video streams of Jabber video
Ingress calls marked AF42
5xxx AF42 Policy
1 • Clients enabled for RTCP
5061 ACL CS3
• Video endpoint and IP phone
classification based on
EF
conditional trust and CDP:
Trust • Audio streams of all calls (voice-
AF41 only and video) marked EF
• Video streams of video calls
CS3 marked AF41
• All Video endpoints enabled for
RTCP
WAN Ingress QoS Marking Policy
Ingress
Policy
1 Router (config-if)# service-policy input INGRESS-MARKING
! Attaches service policy to interface
! This section configures the ACL’s 2

access-list 102 permit tcp any any range 5060 5061
! This section configures the policy-map to set DSCP

for Trusted and Untrusted Voice, Video and SIP Signaling
on ingress
! This section configures the classes
class-map match-any VOICE policy-map INGRESS-MARKING
match dscp ef class VOICE
set dscp ef
class-map match-any PRIORITIZED-VIDEO
match dscp af41 class PRIORITIZED-VIDEO
class-map match-any JABBER-VIDEO set dscp af41
match dscp af42
class JABBER-VIDEO
class-map match-any SIGNALING-SIP set dscp af42
match dscp cs3 class SIGNALING-SIP
3 set dscp cs3
class class-default
Ingress
Policy


for Trusted and Untrusted Voice, Video and SIP Signaling
on ingress
class-map match-any VOICE 4 policy-map INGRESS-MARKING
set dscp ef
class-map match-any PRIORITIZED-VIDEO
match dscp af42
match access-group 101 class JABBER-VIDEO
match dscp cs3
class SIGNALING-SIP
3 set dscp cs3
class class-default
Ingress
Policy

5 for Trusted and Untrusted Voice, Video and SIP Signaling

on ingress
class-map match-any VOICE 4 policy-map INGRESS-MARKING
set dscp ef
class-map match-any PRIORITIZED-VIDEO Egress
Policy
6
match dscp af42
class JABBER-VIDEO
match dscp cs3 class SIGNALING-SIP
3 set dscp cs3
class class-default
Scheduling AF41 WRED thresholds
(i.e., drop AF41 last)
Audio of
IP Phones EF
 Audio/EF is mapped to the priority
EF PQ
queue
Audio of
EF – Includes audio streams from video
Video endpoints and Jabber clients (voice-

Audio of EF only and video calls)
Jabber
 Video streams of video calls (AF41)
Video of
and video streams of Jabber calls
Video AF41 Video (AF42) are mapped to the same
CBWFQ CBWFQ
Video of
 WRED is configured on the video
Jabber AF42 queue:
– min-max thresholds for AF42:
queues
~10% - ~30% of queue limit

other

(i.e., drop AF42 first) – min-max thresholds for AF41:
~45% - 100% of queue limit
WAN Egress Queuing Policy UDP Ports DSCP
3xxx EF
7
5xxx AF42 Egress
Policy 6
5061 ACL CS3
WAN
EF
Trust
! This section applies the policy-map to the Interface AF41
Egress
Policy
6 Router (config-if)# service-policy output EGRESS-QUEUING CS3
! This section configures the bandwidth for

all collab traffic
…
WRED Configuration Example: 10 Mbps WAN Interface
100%
Drop Probability
policy-map WAN-EDGE
class VIDEO
~ ~ bandwidth percent 40
random-detect dscp-based
~
mark-probability-denominator
5%
WRED AF42
mark 2%
probability AF41
(100/denom)
30 80 120 256
WRED
min threshold Queue Occupancy (packets)
WRED
max threshold
Provisioning and Resource Control
Admission Control
• Priority queue is over-provisioned
EF Voice or protected by ELCAC for audio
Queue from all users at the site
AF41
• Video queue is over-provisioned
for video room and desktop
Video endpoints:
Queue WAN • Usage ratios are applied to desktop
AF42 Link video endpoints
• Jabber video streams can use any
bandwidth unused by video room
systems
other • During congestion, Jabber video
voice-only call
queues streams are subject to WRED drops

and dynamically reduce video bitrate
• Audio is not impacted by drops
Provisioning – Central Site
• Voice queue (PQ): 10 Mbps (L3 bandwidth)
500 users:  125 calls @ G.711/G.722
500 Jabber clients Voice
300 IP phones 10 Mbps
30 video endpoints
• Unified CM Location link bandwidth for the
(room + desktop) voice pool:
1 immersive telepresence
2 Telepresence Servers Video WAN
125 * 80kbps = 10,000 kbps
55 Mbps
Link
• Video queue: 55 Mbps (L3 bandwidth)
voice 100
10% Mbps • Immersive Endpoint: 2 Mbps * 1 call = 2 Mbps
• Video endpoints: 1.2 Mbps * 30 calls * 0.2 = 7.2
Default
31 Mbps
Mbps
video
55% • TelePresence Servers: 1.5 Mbps * 40 calls * 0.5
= 30 Mbps
• 55 Mbps - (2 Mbps + 7.2 Mbps + 30 Mbps) =
15.8 Mbps for Jabber Media
 18 Jabber Video calls @ 576p or 50 @ 288p
(plus any leftover bandwidth)
Provisioning – Large Branch
• Voice queue (PQ): 3.4 Mbps (L3
50 users: bandwidth)
Voice
50 Jabber clients
30 IP phones
3.4 Mbps  42 calls @ G.711/G.722
6 video endpoints
(room + desktop)
• Unified Location link bandwidth for the
Video WAN
18.7 Mbps
Link
voice pool:
42 * 80 Kbps = 3.360 Mbps
voice 34
10%
Mbps
• Video queue: 18.7 Mbps (L3
video
Default
10.5 Mbps
bandwidth)
55%
• Video endpoints: 1.2 Mbps * 6 calls = 7.2 Mbps
• 18.7 Mbps - 7.2 Mbps = 11.5 Mbps for Jabber
Media
 13 Jabber Video calls @ 576p or 36 @ 288p
Provisioning – Small Branch
• Voice queue (PQ): 2 Mbps (L3
15 users: Voice
2 Mbps
bandwidth)
15 Jabber clients
10 IP phones
 25 calls @ G.711/G.722
2 video endpoints
(room) Video WAN
Link
4 Mbps
voice pool:
10
Mbps 25 * 80 Kbps = 2 Mbps
voice
20%
video
Default
3.6 Mbps
• Video queue: 18.7 Mbps (L3
40%
bandwidth)
• Video endpoints: 1.2 Mbps * 2 calls = 2.4 Mbps
• 4 Mbps – 2.4 Mbps = 1.6 Mbps for Jabber
Media
 2 Jabber video calls @ 576p or 5 @ 288p
Provisioning – Micro Branch
5 users: • Broadband Internet connectivity +

Voice
5 Jabber clients 1.25 Mbps DMVPN to central site
3 IP phones
1 video endpoint WAN
Video
1.5 Mbps Link • Configure interface of VPN router to
5
match broadband uplink speed
Mbps
Internet Default • Enable QoS on VPN router to
2 Mbps
prevent bufferbloat* from TCP flows
DMVPN
Tunnel Broadband
Modem/Router
• Asymmetric download/upload
QoS-enabled
Router
voice
25%
broadband: consider limiting
video
30%
transmit bitrate on video endpoint
*: see reference slide #29 in this session

CAC for Video Considerations
• In branch sites with lower-speed WAN links over-provisioning the video queue is
not feasible
• E-LCAC will be applied also to video streams in these sites
• Use site-specific region configuration to limit maximum bandwidth used by video
endpoints and Jabber clients
• Device mobility needed if Jabber users roam across sites
Provisioning – Large Branch with Constrained WAN Link (CAC for video)
• Voice queue (PQ): 2 Mbps (L3 bandwidth)
50 users:
Voice
 25 calls @ G.711/G.722
50 Jabber clients 2 Mbps
30 IP phones
4 video endpoints
(room + desktop)
Video WAN voice pool:
Link
4 Mbps
25 * 80 Kbps = 2 Mbps
10
voice
Mbps • Video queue: 4 Mbps (L3 bandwidth)
20% Default • Possible breakdown:
video 3.6 Mbps
40% 2 calls @ 576p (768 kbps) +
5 calls @ 288p (320 kbps) = 3136 Kbps
• CUCM Location link bandwidth for video
calls: 3.2 Mbps (L3 bandwidth)
• Leaves room for L2 overhead, burstiness and
Jabber audio-only calls marked AF41
Conclusion
• Combine QoS tools, media resilience and dynamic adaptation to build a self-
regulating system that makes optimal use of available network resources
• Use CAC when and where needed
• When managing bandwidth with Media Resilience and Rate Adaptation techniques is
not an option (i.e. extreme contention on WAN bandwidth)
Future:
• Leverage SDN approach to simplify QoS deployment and enable dynamic QoS
policies for media from untrusted devices
• Enable endpoints and network devices to exchange information about media
and network status along the path (e.g., relative priority between sub-streams,
network congestion status, ...)
136
Continue the Conversation using Cisco Spark
Use Cisco Spark to continue the conversation or ask any additional questions
with the speaker for this session.
1. Sign up at http://www.ciscospark.com/
2. Download the App from iOS App Store, Google Play Store, or from
http://download.ciscospark.com/
3. Send an email to spark-at-ciscolive@cisco.com with the message “Please
add me to the BRKCOL-2616 room”
Visit the World of Solutions Cisco Spark area for demos

Thank you

Brkcol 2616

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brkcol 2616

Uploaded by

Copyright:

Available Formats

QoS Strategies and Smart Media

Techniques for Collaboration Deployments

Stefano Giorcelli Glen Lavers

Mobile, software endpoints

On-premise Call Control

Remote Sites Home/Mobile Users

• Deploying “pervasive” video is challenging

• Cloud services, B2B/B2C communications and remote/mobile access make it

 Video streams are bursty and adaptive, audio

... ... R2 1011010010

Leverage media resilience and

3 the network and mark their

• Separate UDP port ranges for video

Audio stream CS4 Real-Time

Desktop video Telepresence

Audio stream EF “Opportunistic”

BW Assigned to LLQ Classes

Video of Jabber AF42

AF42 WRED thresholds

• Identification and Classification

• Mobile and Remote Access

VIDEO Bandwidth  Bandwidth:

L1/L2 Transport Issues

Out of Sync (OOS)

 Loss of a P-frame triggers request for a new I-frame

 Delay seen by 1 audio stream

 Delay seen by 1 audio stream

LTRF with Repair

Optimize use of available

 Each frame must be

 Keep encoder and decoder in sync with active feedback messages

LTRF Binary XOR R1 FEC

 Two approaches possible:

Choose starting bitrate

no Packet loss > Ld% Packet loss < Li% no

Decrease bitrate by Increase bitrate

MCU (4.5) (4.5)

• Media resilience mechanisms help mitigate impact of video traffic on the

• Dynamic rate adaptation creates an opportunity for more flexible provisioning

IDENTIFICATION & CLASSIFICATION

• Trusted devices • Map identified • Low Latency • Bandwidth • Prime Collaboration

• Weighted Random BRKCOL-2320 - Cisco

Broadcast Video CS5 (Optional) PQ Cisco IP Video Surveillance / Cisco Enterprise TV

Realtime Interactive CS4 (Optional) PQ Cisco TelePresence (Legacy CTS-series)

Network Control CS6 BW Queue EIGRP, OSPF, BGP, HSRP, IKE

Signaling CS3 BW Queue SIP, H.323, SCCP

Ops / Admin / Mgmt (OAM) CS2 BW Queue SNMP, SSH, Syslog

Best Effort DF Default Queue + RED Default Class

[MLS] mls qos trust dscp

[MLS] no mls qos trust

No Trust [MQC] no qos trust dscp

CoS-to-DSCP Mapping Table

mls qos map cos-dscp 0 8 16 24 32 40 48 56

Full Trust IP WAN

priority percent 10 class-map match-all VOICE

! This section applies the policy-map to the Interface

! This section configures the bandwidth for all collab

CDP Support required

Trusted Endpoints / Devices

Untrusted Endpoints / Devices

Trusted Devices WiFi

UCM IM&P VCS / MCU/TS 49

DSCP for Audio Calls