10 - Traffic Engineering Flashcards
The process of re-configuring the network in response to changing traffic loads, to achieve some operational goal
Traffic engineering
network operator might want to reconfigure the network in response to changing traffic loads
1) to maintain traffic ratios in a peering relationship
2) to relieve congestion on certain links in the network
3) to balance load more evenly across the available links in the network
How should routing adapt to traffic
1) Avoid congested links
2) Satisfy application requirements
Traffic engineering has 3 steps
1) Measure
2) Model
4) Control
Intradomain Traffic Engineering:Tuning link weights
1) Routers flood information to learn topology
2) Operator configures link weight
Ways that operators can set the links
1) Inversely proportional to capacity
2) Proportional to propagation delay
3) Network wide optimization
Intradomain Traffic Engineering: Optimization
Input: Graph(R,L)
Output: Set of link weights (WL)
Utilization
The amount of traffic on the link divided by the capacity
In practice, we have these operational realities to worry about (Intra-domain routing)
1) Minimizing the number of changes to the network
2) Resistant to failure
3) Robust to measurement noise
Intradomain
Interdomain
Intradomain: within a domain
Interdomain: between domains
Examples of Interdomain routing
1) Peering between 2 ISPs
2) Peering between a university network & its ISP
3) Peering at an Internet exchange point
4) Routing across multiple datacenters
Interdomain routing or Interdomain traffic engineering
Routing between ASes
- Alleviating congestion on edge links - Using new/upgraded edge links - Changing end to end path
Goals for Interdomain Traffic Engineering
1) Predictability
2) Limit influence of neighbors
3) Reduce overhead of routing changes
Equal cost multipath (ECMP)
1) Intradomain routing - Equal cost multipath (ECMP)
How can a source router adjust paths?
1) Alternating between forwarding table entries.
By having multiple forwarding table entries for the same destination and splitting traffic flows across the the multiple next hops depending on the hash of th IP packet header.
Data Center Networking
Characteristics
1) Multi-tenacy
2) Elastic resources
3) Flexible service management
Data center networking challenges
1) Traffic load balance
2) Support for virtual machine mugration
3) Power savings
4) Provisioning
5) Security
Data center topology
Internet Core Aggregation Access ->Servers
Scale problem
10s of thousands of servers on a flat layer-2 topology where all of the servers have a topology independant MAC hardware address and in default every switch in topology has to store a forwarding table entry for every MAC address (Huge forwarding tables!)
Scale problem solution
Pseudo Mac -> real MAC
Fabric manager is involved here.
VL2 (Valiant Load Balance)
Goals
1) Spread traffic
2) Location independence
Jellyfish
Networking data centers randomly
Jellyfish goals
Goals:
High throughput -> Big data
Incremental expandability -> easy replacement of servers
Problems:
Structure constraints expansions
Where does DC structure constrain expansion
Top-level switches
Jellyfish’s topology
Random regular graph
Jellyfish construction (RRG -> random Regular Graph)
N(ki-ri) servers
ri ->switches
ki->ports
N -> racks
Random Regular Graph
RRG(N,k,r)
Constructing Jellyfish
1) Pick a random switch pair
Open questions on Jellyfish topology
Topology Design How close are random graphs to optimal What about heterogeneous switches System Design Cabling Routing