10. Traffic Engineering Flashcards
Traffic Engineering
Process of reconfiguring the network in response to changing traffic loads, to achieve some operational goal
Why traffic load? Operator might want to ____
Peer ratios
relieve congestion
balance load
Self managing ip networkings
TCP
Routing
Self managing ip networkings: Problem
May not run efficiently
Intradomain TE: Tuning link weights
intradomain: routers flood information to learn topology
operator can configure the weights on links to help decide the flow of that information through the network to help avoid congestion
Traffic Engineering 3 steps
Measuring the network
Forming a model
Reconfiguring the network to assert control
Measure
Topology
Traffic
Intradomain TE: Optimization
Input: Graph G(R, L)
Output: Set of link weights wl
Graph G(R, L) variable meanings
R = routers L = links
cl
capacity of l
Traffic matrix Mij
traffic from router i to router j
Objective function
Objective: minimize f(ul/cl)
Utilization function
ul/cl
traffic on link/capacity
Intradomain routing
within a domain (ISPs, campus, datacenter)
interdomain routing
between networks
interdomain traffic engineering
Using new/upgraded edge links
Alleviating congestion on edge links
cloning end to end path
Goals for interdomain traffic engineering
Predictability
Limit influence of neighbors
Reduce overload of routing changes
How to be predictable
No globally visible changes
How to limit influence of neighbors
consistent adverts, limit AS path
How to reduce overhead of routing changes
group prefixes
multipath routing
operator can establish multiple paths in advance
traffic splitting
change % of traffic sent on path
Data center networking Characteristics
multi tenancy
elastic resources
flexible service management
DAta center networking challenges
traffic load balance support for virtual machine migration power savings provisioning the network when changes occur security
Data Center topology
- Each pod has address
2. Each server has a psuedo mac corresponding to pod
Data center TE
Limited server to server capacity - links at the top of a fat tree topology are over subscribed
Fragmentation - significant’y lowering utilization
Valiant Level Balance goals
- Spread Traffic
2. Location independence
Jellyfsh
Networking data centers randomly
Jellyfish Goals
High throughput (Big data) incremental expandability (easy replacement of servers)
Jellyfish problems
Structure constrains expansion
Jellyfish topology is called
randomly regular graph
randomly regular graph
randomly selected from regular graphs
regular graph
each node has the same degree
each switch is a node
Constructing a Jellyfish topology
- Pick a random switch pair where they have a free port to become neighbors
- join with link.
- repeat until no further links can be added
