Enabling Ultra Low Latency Applications Over Ethernet
-
Upload
kaseem-richard -
Category
Documents
-
view
47 -
download
1
description
Transcript of Enabling Ultra Low Latency Applications Over Ethernet
Enabling Ultra Low Latency Applications Over Ethernet
Grant Kirkwood
Chief Technology Officer
Mzima Networks
Services are fast becoming packet-basedResidential
Circuit-switched voiceand
broadcast video
Circuit-switched voiceand
broadcast video
Digital broadcastIPTV, VoIP
Internet videogaming
Digital broadcastIPTV, VoIP
Internet videogaming
Enterprise
Dedicated connectionsT1/T3, FR, ATM
Digital PBX
Dedicated connectionsT1/T3, FR, ATM
Digital PBX
Ethernet ServicesVirtual circuits
IP PBXVoIP
Ethernet ServicesVirtual circuits
IP PBXVoIP
Wireless
Voice optimized2G network
Voice optimized2G network
Data optimizedVideo enabled3G/4G/WiMax
Data optimizedVideo enabled3G/4G/WiMax
Services are stream-based, not packet-based
Digital broadcastIPTV, VoIP
Internet videogaming
Digital broadcastIPTV, VoIP
Internet videogaming
Ethernet ServicesVirtual circuits
IP PBXVoIP
Ethernet ServicesVirtual circuits
IP PBXVoIP
Data optimizedVideo enabled3G/4G/WiMax
Data optimizedVideo enabled3G/4G/WiMax
Services are stream-based, not packet-based
Digital broadcastIPTV, VoIP
Internet videogaming
Digital broadcastIPTV, VoIP
Internet videogaming
Ethernet ServicesVirtual circuits
IP PBXVoIP
Ethernet ServicesVirtual circuits
IP PBXVoIP
Data optimizedVideo enabled3G/4G/WiMax
Data optimizedVideo enabled3G/4G/WiMax
Voice, video and other services – streams of data
Internet Protocol (IP) relies on small packets, not streams of data
IP Networks designed to carry
“best effort” traffic
Email, web browsing, “instant” messaging….
Applications are not time-sensitive.
Networks designed to carrypacket-based data are now being asked
to carry streaming data
IP Networks designed to carry“best effort” traffic
Email, web browsing, “instant” messaging….
Applications are not time-sensitive.
Networks designed to carrypacket-based data are now being asked
to carry streaming data
Enterprise
Penetration of Ethernet, fiber, collaboration, and interactive video
Increased workforce mobility requiring seamless access
Maturing virtualization, cloud-based applications & telepresence
Consumer
Maturity of multiservice offerings Increased mobile data penetration
and improved ease of use Growing adoption of online video
and Web 2.0 applications
Perceived value has shifted from service provider to application provider
New applications created daily
Applications become more and more sensitive to network conditions
QoS policies are being created to support these technologies
An increasing number of QoS policies are creating challenges for service providers
New applications created daily
Applications become more and more sensitive to network conditions
QoS policies are being created to support these technologies
An increasing number of QoS policies are creating challenges for service providers
But revenue growth is slowing
2004 2006 2008 2010
10%
20%
30%
41%
3%7%
13%13%
-5%-3%-1%4%
Broadband
Wireless
Wireline
**
** Sources: Yankee Group and Pyramid Research
Traffic keeps growing …
66 Tbps
2007
134 Tbps
2011
Consumer
Enterprise
56 Tbps
101 Tbps
33 Tbps
*
* Source: McKinsey & Company
10 Tbps
Service provider realities
How do we deliver
real-time services with carrier-grade
QoS?
How do we deliver
real-time services with carrier-grade
QoS?
QoS Model
Ethernet Performance Metrics
• Packet Loss• Latency• Jitter
• Packet Loss• Latency• Jitter
Ethernet Performance Metrics
Packet LossPacket Loss
LatencyLatency
JitterJitter
Ethernet Performance Metrics
Packet LossPacket Loss
LatencyLatency
JitterJitter
Ethernet Performance Metrics
Different applications are sensitive to different performance metrics
Different applications are sensitive to different performance metrics
Class/Quality of Service
QoS/CoSQoS/CoS
QoS Model
QoS Model
Methods to separatetraffic into different “buckets”
or QoS policies
Methods to separatetraffic into different “buckets”
or QoS policies
QoS Model
“FIFO” is not intelligent ordering
Order by priority
“FIFO” is not intelligent ordering
Order by priority
QoS Model
Drop packets for applications not
sensitive to packet loss
Drop packets for applications not
sensitive to packet loss
Non-stream services
Non-stream services
QoS Model
Size throughput to availability
Limit/buffer traffic that won’t be impacted by policing
Size throughput to availability
Limit/buffer traffic that won’t be impacted by policing
QoS Model
QoS Model
Policing vs Classifying
• Bandwidth policing is “dumb”
• Metering drops packets without discretion
• Congestion causes buffering (on routers)
• Buffering causes latency
• Each application sensitive to specific metrics
• Key is accurate classification
End to end QoS
VoIPVoIP
GoldGold
SilverSilver
Best EffortBest Effort
VoIPVoIPGoldGold
Classification/Policing
SilverSilver
Schedule
VoIPVoIPGoldGoldSilverSilver
Transmit
End to end QoS
• End-to-end QoS requires a technology that exists end-to-end
• Difficult to achieve in multi-vendor, multi-carrier or multi-technology networks
• New technologies are being developed to address this limitation
Mzima Case Study
• Problem: provide true Carrier Ethernet
• Differentiate product from other carriers
• Provide End-to-End QoS
• Carry carrier-grade Services
• Application-level granularity of QoS profiles
• Meet requirements of true Carrier Ethernet
Mzima Case Study
Mzima Case Study
• VLAN bridging or tunneling
• Layer 2 MPLS
• Layer 3 MPLS
• PBB-TE
PBB-TE
• Provider Backbone Bridge – Traffic Engineering
• End-to-End QoS
• Stringent performance metrics
• First Carrier Ethernet protocol not integrated into a layer 3 protocol (MPLS)
PBB-TE BEB
PBB-TE BEB
PBB-TE BCBs
PBB-TE with Y.1731 Performance Management• Performance Management between Tunnel Endpoints
• Provides Service Independent Tunnel Monitoring• Enhanced Scalability as 1,000’s of services may traverse the
tunnel without the need to monitor every service• Leverages 802.1ag frames for reduced overhead
• Multiple packets sent at 100ms interval to perform the test• Frame Delay / Frame Delay Variation / Loss Measurement• 2-way Delay Roundtrip Measurement• 1-way Delay Measurement (requires common time base)• Single Ended Frame-Loss (MEP to MEP)
PRIMARY
BACKUP
Y.1731 Performance Management
PBB-TE Tunnel Performance Management
Management Plane: Y.1731Round trip delay/jitter and single ended frame loss (MEP to MEP)
Non-Service Affecting Utilizes IEEE 802.1ag format frames for test packets Unicast messages to a specific MEP Multiple packets sent at 100ms interval to perform the test Delay, Jitter, and Frame Loss measurements Test results remain until the next test is run or until reboot of switch MIPs do not participate in delay/jitter/frame loss measurements
MEP 10
MEP 12
MEP 11
MIP
MIPMIP
802.1ag CCMs
Thank You