Netscope: Traffic Engineering for IP Networks

Download Report

Transcript Netscope: Traffic Engineering for IP Networks

Netscope: Traffic Engineering for IP
Networks
A.Feldmann, A.Greenberg, C.Lund,
N.Reingold & J.Rexford
AT&T Labs
presented by
Rajendraprasad B.Hosamani
ECE Dept, Umass, Amherst
7/17/2015
Netscope
1
Organization
•
•
•
•
•
Motivation
Netscope tool overview
Building blocks Netscope
Internet structure and building blocks
Unleashing the power of Netscope
– performing “what-if” experiments
• Conclusion
7/17/2015
Netscope
2
Motivation
• Today's IP networks are extremely large
• Understanding them requires one to understand
– Routing policies
– Current traffic flows
– Network configuration
• Managing such an IP network requires one to perform
traffic measurement and network modeling
• State-of-the-art methods rely on combination of intuition,
experimentation, trial and error
• Netscope attempts to provide s/w systems to do the same
7/17/2015
Netscope
3
Why do we need such tools
• Service Quality
– ISP’s provide performance guarantees in the form of SLA’s
– Customers monitor the compliance.
• Interdependent tunable parameters
• Network growth
• Traffic variability
– Study the fluctuations in the distributions of IP traffic
7/17/2015
Netscope
4
What is Netscope
• Netscope is a unified set of software tools for
managing the performance of IP backbone
networks.
• With Netscope one can
– generate global views of the network
– Visualize network-wide implications of local changes
– Experiment with changes in network configuration in a
simulated environment
– perform performance debugging
7/17/2015
Netscope
5
Building Blocks of such a tool
• Topology deduction.
• Traffic measurements.
• Combine diverse network configuration and traffic
measurements into a joint model.
• Some way to capture the routing behavior in the
network.
• Visualization of all the above data in a way that it
makes easy sense.
7/17/2015
Netscope
6
Components of Netscope toolkit
Visualization
Netscope
Routing Model
Data Model
Configuration
7/17/2015
Measurements
Netscope
7
A Note on implementation
• Architectural/Technological changes can occur
• Higher level modules must be flexible to handle
this
• Topology and traffic data can be derived from
variety of sources
– Lower level mainly consists of raw parsers
– Should be designed for simplicity and extensibility
7/17/2015
Netscope
8
Structure of Today’s Internet
Devices & Links
•Access Routers/Links
•Peering Links
•Backbone Routers/Links
•Gateway Routers
•Trunks/Devices
Policies & Logical Structure
•AS and Areas
•Internal/External policies(BGP)
•OSPF/IS-IS/RIP
7/17/2015
Netscope
9
Links Vs Trunks
Link
RS
L2R
L21
7/17/2015
Trunk
Devices
Netscope
RD
L22
10
Topology Model
• Unidirectional Link Attributes
– Router originating link,name of the router card,IP address of the
interface, description of the purpose, capacity, OSPF weight
• Bi-directional Link Attributes
– Same as above, with some attributes common for both directions.
For example: OSPF area
• Router Attributes
– name, loopback IP address, type of router (AR, BR, IGR), location
– The links which the router originates
7/17/2015
Netscope
11
Topology Model (Contd.. ..)
• Device Attributes
– name, location, trunks which originate at the device
• Trunk Attributes
– Links traversing a given trunk.
• Link Attributes
– name,Source/Destination routers, Trunks making the link
7/17/2015
Netscope
12
Traffic and what we are interested in
• Some Observations
– Customers connect to backbone via multiple access links
• Hence traffic introduced by a customer should be modeled as a
demand from an access link to a set of peering links
– Many external addresses are reachable via multiple peering links
• Hence traffic from external internet to a customer should be modeled
as a demand from a peering link to a set of access links.
7/17/2015
Netscope
13
Traffic (contd.....)
7/17/2015
Netscope
14
Determining Traffic Demands
• Determining Access links associated with a customer
– Based on forwarding table at each access router
– Each table entry consists of customer prefix,card name of outgoing
link
– Router configuration files associates prefixes to links
• Determining Peer links associated with a customer
– Make use of the BGP routing tables which consists of next-hop AS
path and IGR
– Each external prefix is associated with a set of peering links
– IGR and next hop gives us information about the peering links
7/17/2015
Netscope
15
Traffic Measurement
• Flow level measurement done at the edge of the network
• Measurement done by dumping the main IP, TCP/UDP
header fields ,number of packets transmitted, bytes
transferred, start and finish time of the flow.
• The source and destination IP addresses of the flow can be
associated with the appropriate prefix, and matched to the
corresponding access/peering links.
• With this aggregate traffic due to a customer
(entering/leaving the network) can be calculated.
7/17/2015
Netscope
16
Routing Model
• Netscope’s routing module determines path(s) chosen by
OSPF for each traffic demand.
• Netscope considers a single instance of the network
topology and OSPF configuration
• Netscope does not implement route summarization
7/17/2015
Netscope
17
OSPF Tie-Breaking
• There exists multiple shortest paths between a pair of
routers. Example :
– Due to parallel links for additional capacity,similar OSPF weights.
• Traditionally hashing is used to break ties
• Hashing is vendor specific
• Hence Netscope implements a hashing function which
splits the traffic evenly.
7/17/2015
Netscope
18
OSPF Tie-Breaking
7/17/2015
Netscope
19
Visualization
•Visualization environment provides
a way to examine various network
parameters and objects
•Figure displays attributes of a router
object which starts at a router in Dallas
and terminates at a router in LA.
7/17/2015
Netscope
20
Visualization (contd.....)
• Netscope allows to monitor/visualize a large number of
statistics
7/17/2015
Netscope
21
Features of Netscope
• With Netscope one can limit the display to a few chosen
demands.
7/17/2015
Netscope
22
Unleashing the power of Netscope
Low utilization (<= 30%)
Medium utilization ( 30%< u <= 60%)
High utilization (> 60%)
7/17/2015
Netscope
23
Decreasing OSPF weights
• Experimentation can be done
in a similar way to divert some
of the traffic on the link
between Cambridge and Chicago
7/17/2015
Netscope
24
Conclusions and Future work
• Netscope is a powerful tool for network
visualization
• Aids in performance debugging.
• Netscope works on a static feed of topology
– Future work may enhance it to continuous feed
of topology
• Support for wireless Ad-hoc networks??
7/17/2015
Netscope
25
Thank you!!!!!
7/17/2015
Netscope
26