What is Traffic Engineering
The process of reconfiguring the network in response to changing traffic loads, to achieve some operational goal
What are some of the goals a network operator would hope to achieve with Traffic Engineering?
- Maintain Traffic ratios in a peering relationship
- Relieve congestion
- Balance load more evenly across available links in the network
What are the questions which Traffic Engineering attempts to answer?
Does the network run efficiently?
How should routing adapt to traffic?
- Avoid congested links
- Satisfy application requirements
How can an operator affect how traffic flows in an Intradomain TE topology?
Configure the link weights in one of three ways:
- Inversely proportionate to capacity
- Proportional to propagation delay
- Network-wide optimization
What are the three steps to traffic engineering?
- Measure the network, to determine the current traffic loads
- Forming a model of how configuration affects the underlying paths in the network
- Reconfigure the network to exert “control” of how traffic flows
The Intradomain traffic engineering problem
Input 1: Graph G(R,L), with R = set of routers, L = set of unidirectional links. Each link L has a fixed capacity, c_L
Input 2: Traffic Matrix M_ij, traffic from Router “i” to router “j”
Output: set of link weights, w_L
What si the difference between Intredomain routing and Interdomain routing?
Intradomain: within a domain (e.g. ISP, campus, datacenter)
Interdomain: Between domains
Which of the following are examples of interdomain routing?
A: Perring between two ISPs
B: Peering between a university network and its ISP
C: Peering at an internet exchange point (IXP)
D: Routing in a data center:
E: Routing across multiple data centers
A, B, C, & E
Interdomain Traffic Engineering
- Alleviate congestion on edge links
- Using new/upgraded edge links
- Changing end-to-end path
- It involves the reconfiguration of BGP
Goals for Interdomain TE
- Predictability
- No globally visible changes
- Limit influence of neighbors
- Consistent advertisements, limit AS path
- Reduce overload of routing changes
- Group prefixes together according to those with common AS paths
What is Multipath routing
Where an operator can establish multiple paths in advance. This applies both to Inter and Intradomain routing
How can a source router adjust paths?
A. Dropping packets to cause TCP backoff
B. Alternating between multiple forwarding table entries
C. Sending alerts to incoming senders
B
What characterizes a data center
- Multi-tenancy
- allows provider to advertise cost of shared infrastructure
- Infrastructure must provide security and resource isolation
- Elastic resources - as traffic expands or contracts, so does the data center usage.
- Flexible service management - the ability to move work or workloads to other locations inside the datacenter.
- This workload movement, migration is what introduces the need for traffic engineering solutions
A key enabling technology in data center networking
The ability to virtualize servers. This enables the quick migration and movement of servers
What are the challenges of data center networking?
- Traffic load balance
- Support for virtual machine migration
- Power savings
- Provisioning when demand fluctuates
- Security
What are the three layers of the data center topology?
Core (connected to layer-2)
Aggregation (connects access layer)
Access (connects servers)
how is the “core” layer of a data center connected?
Historically it used a layer 3 topology, but modern centers are all layer-2 topology.
This makes it easier to do migration of services, so they can stay on the same layer 2 network and would not need new IP addresses.However this makes scaling difficult.
What are some problems at the core level of a data center topology?
points of failure
over-subscription - 200x as much traffic as the links at the bottom of the hierarchy
How do you address the problem of “scale” with data center “Flat” typologies?
- Introduce “pods”
- Each pod has it’s own address
- Each server has pseudo MAC corresponding to Pod.
How do you deal with data centers mapping pseudo MACs to real MAC addresses?
Watch 13.13
Where does Jellyfish DC structure constrain expansion
A: Individual servers
B: Aggregation switches
C: Top level switches.
C
What is Jellyfish?
A technique to network data centers randomly.
Goals:
High throughput, -> big data
Incremental expanability ->Easy replacement of servers
Problem:
- hypercube : 2k switches
- 3-level fat tree: 5k^2/4 switches
What are the two main goals of valiant load balancing?
1 spread traffic evenly across the servers
2. Ensure traffic load is balanced independent of the destinations of the traffic flows
