Traffic Engineering
Process of reconfiguring the network in response to changing traffic loads, to achieve some operational goal.
- Peering ratios
- Relieve Congestion
- Balance load
“Self-management”
TCP, routing
How should routing adapt to traffic?
- Avoid congested links
- Satisfy application requirements such as delay
Intradomain TE: Tuning Link Weights
- Routers flood information to learn topology
- Operator configures link weights
- Inversely proportional to capacity
- Proportional to propagation delay
- Network-wide optimization
3 Steps of Traffic Engineering
- Measuring the network to figure out the current traffic loads
- forming a model of how configuration affects the underlying paths on the network
- Reconfiguring the network to exert control over how the traffic flows through the network
Measure, Model, Control
Intradomain
Within a domain (e.g., ISP, campus, datacenter)
Interdomain
between domains
Interdomain Traffic Engineering
- Alleviating congestion on edge links
- Using new/upgraded edge links
- Changing end-to-end path
Reconfiguration of BGP
Goals of Interdomain TE
- Predictability
- > no globally visible changes - Limit influence of neighbors
- > consistent adverts, limit AS path - Reduce overload of routing changes
- > group prefixes
How can source router adjust paths?
Alternating b/w forwarding table entries
Data Center Networking
Characteristics of a data center
1. Multi-tenancy
\+ amortization of cost
- security, resource isolation
2. Elastic resources
3. Flexible service management
--> workload movement, migration
enabler: virtualizationData Center Networking Challenges
- Traffic load balance
- Support for virtual machine migration
- Power savings
- Provisioning
- Security
Data Center TE
- Limited server-to-server capacity
- > links @ “top” of fat tree topology are over-subscribed
- Fragmentation
Goals of Valiant Load Balancing in VL2 network
- Spread traffic (evenly across servers)
2. Traffic load is balanced independently of the destinations of the traffic flows
Jellyfish
Goals:
- High throughput -> big data
- Incremental expandability -> easy replacement of servers
Problem: Structure constrains expansion
