traffic engineering in software-defined networks
TRANSCRIPT
UWB
Traffic Engineering in Software-Defined Networks
Hai Dinh Tuan
Next Generation Networks & Future Internet Technologies Seminar
• Introduction – Traffic Engineeirng & SDN
• What's new with TE in SDN
• New Traffic engineering techniques in SDN
• Conclusion
Agenda
Introduction
Introduction - Traffic Engineering
Traffic Engineering (TE) can be defined as a set of techniques, designed
to manage the allocation of network resources to carry traffic subject
to constraints
The separation of
control plane (routing
functionalities) and
data plane (forwarding
functionalities)
Introduction - SDN
Source: SDNCentral
Traffic Engineering in SDN
1. Future networks require a high performance and highly
flexible infrastructure• Rapid expansion of today Internet
• The growth of mobile devices, as well as real-time data
2. New traffic engineering techniques must face
unprecedented challenges• Old periodic data sampling mechanism can create a significant
overhead.
Future networks requires new techniques
1. Simplified network management• Controller has the global view of network
• Focus on specifying network-level objectives
2. Flexibility• Eliminate vendor lock-in situation
• New network features can be easily developed and deployed
3. Reliability• Programmability makes network behavior more deterministic
SDN makes TE easier
4. Higher Efficiency• Organizations can develop their own algorithms to achieve
optimal performance
• Controller can be regularly upgraded with the latest hardware
technologies
5. Innovation support• Develop new function for controller without modifying switches
and vice versa
• Enables external contributions
SDN makes TE easier (2)
Specific techniques
Research directions in TE for SDN
1. Efficient flow managementWhen there is a high number of incoming flow, overload may occur
2. Fast recovery mechanismController must recalculate new routes and notify all related
switches in case of network failure
3. ConsistencyGuarantee consistency during network updates
4. Low-overhead traffic analysisSDN requires new network management mechanisms
Research directions in TE for SDN
1. Efficient flow managementWhen there is a high number of incoming flow, overload may occur
2. Fast recovery mechanismController must recalculate new routes and notify all related
switches in case of network failure
3. ConsistencyGurantee consistency during network updates
4. Low-overhead traffic analysisSDN requires new network management mechanisms
How to avoid overload when there are many new flows are injected?
Flow management
1. Switch load-balancingEqual Cost Multi-Path Routing (ECMP)
3. Controller load-balancingLogically distributed controller
Physically distributed controller
Hierarchical controller
Multi-threaded controllers
4. Using multiple flow tablesAlready used from OpenFlow 1.1
DIFANE
Controller should
generate rules, but
don't need to be
involved in the real-
time handling of data
packets
The switches handle
all packets in the data
plane DIFANE flow management architecture [1]
Research directions in TE for SDN
1. Efficient flow managementWhen there is a high number of incoming flow, overload may occur
2. Fast recovery mechanismController must recalculate new routes and notify all related
switches in case of network failure
3. ConsistencyGurantee consistency during network updates
4. Low-overhead traffic analysisSDN requires new network management mechanisms
What is the best way to implement a fault tolerance mechanism?
Fault tolerance
1. Fault tolerance for data planeData plane restoration (reactive)
Data plane protection (proactive)
3. Fault tolerance for control planeBackup controller
CPRecovery
CPRecovery controller backup mechanism [2]
A component runs on top of the
network OS, support Primary-
Backup mechanism.
Switch uses inactivity probes to
check Primary controller. If its
down, switch initiates
connection to secondary
controller.
Research directions in TE for SDN
1. Efficient flow managementWhen there is a high number of incoming flow, overload may occur
2. Fast recovery mechanismController must recalculate new routes and notify all related
switches in case of network failure
3. ConsistencyGurantee consistency during network updates
4. Low-overhead traffic analysisSDN requires new network management mechanisms
How to enforce policies during a network change?
Consistency during network updates
1. Per-packet consistencyEach packet is forwarded through the network according to only
one policy, not the mix of old and new policies
3. Per-flow consistencyAll packets in the same flow will be processed by the same network
configuration
Using configuration version
• Ingress switch marks a configuration version on each
incoming packet
• Packets are processed based on their version value.
• Ingress switch update configuration version when there is
update
• When all packets belong to old policy has left the
network, controller removes old rules from all switches
Research directions in TE for SDN
1. Efficient flow managementWhen there is a high number of incoming flow, overload may occur
2. Fast recovery mechanismController must recalculate new routes and notify all related
switches in case of network failure
3. ConsistencyGurantee consistency during network updates
4. Low-overhead traffic analysisSDN requires new network management mechanisms
How to develop network management tools suit better to SDN?
Traffic Analysis
1. Monitoring frameworkProvides support for network management applications
2. Network invariants checkMake sure that every rule insert/delete operation doesn't lead to net
work inconsistency
3. Debugging toolsIt is possible to debug a whole SDN network as a software
PayLess
PayLess provides a RESTful API
for developing monitoring
applications. They can be
written in any language and can
decide what, where and when to
collect statistics
PayLess (2)
Adjust the monitoring frequency according to network load
by adaptive monitoring algorithm:• Assign a monitoring time out to each flow
• Query the switch(es) for flow statistics when timeout expires
• If no significant traffic change, increase the timeout
• If change in traffic is significant, decrease the timeout
=> Flows that significantly contribute to link utilization has a higher
polling frequency
Conclusion
Conclusion
Conclusion
1. Challenges go hand-in-hand with oppotunitiesThis paradigm shift has many advatages but also requires new
design for TE tools.
2. Many unsolved problemsHybrid SDN network?
Incremental implementation of SDN?
3. Requires collaboration between academia and industry
UWBTHANK YOUHai Dinh Tuan
Next Generation Networks & Future Internet Technologies Seminar