Title: SEQUIN An MPLS Monitoring and Analysis Tool for IP Networks
1SEQUINAn MPLS Monitoring and Analysis Tool for
IP Networks
- Marina Thottan1 , Ken Swanson1 , Michael Cantone1
, - Jennifer Ren2 , Tin Kam Ho3
- 1 Networking Software Department,
- 2 Network and Service Management Department,
- 3 Scientific Computing Department, Bell
Laboratories.
2What is the problem being addressed?
Customer A HQ (NY)
Customer BHQ (NY)
Customer B Branch (Boston)
Gold
- MPLS-based VPN services are being offered by
Service Providers - MPLS tunnels carry multiple classes of traffic
for VPN customers - MPLS network need to be monitored for
- Computing MPLS tunnel utilization
- Detection of Fault/Congestion
- Validation of
- SLA
- Better monitoring implies
- Lower OPEX
- Higher ROI
- Additional Revenue
Silver
Access Point
Bronze
STN-1
STN-4
LSP-1 (Red)
Optical Core
LSP-2 (Blue)
STN-3
STN-2
Customer C HQ (Chicago)
Customer A Branch (Dallas)
Customer CBranch (LA)
3Overview
- SEQUIN is a software tool for online performance
monitoring and analysis of MPLS paths in IP
networks - monitors traffic statistics at edge and
coredevices - synthesizes monitored traffic statisticsto
obtain network-wide QoS statusand general
network element health - continuously displays QoS status andgeneral
network health - easily integrates with higher levelapplications
(e.g., provisioning ,SLAverification, billing) - Assumptions
- DiffServ used to aggregate and classifytraffic
entering at the network edges - MPLS used to provide QoS-guaranteed paths in the
core - SNMP used to collect network device statistics
relevant to QoS
4Innovations and Research Contributions
- Innovations
- Use of near real-time measurements to provide
online view of edge-to-edge QoS status - Online synthesis of edge-to-edge QoS metrics
- Time-synchronized polling and correlation of MIB
data provides accurate synthesis of QoS
information - Online multi-variate predictive algorithm
provides indicators of traffic anomalies - Visualization of network QoS
- Research contributions
- Algorithms for computing MPLS path level QoS
metrics from raw MIB data - Evaluation of SNMP as a scalable network
measurement technique - Algorithms for distributed data aggregation to
minimize the impact of management on network
traffic
5SEQUIN Architecture
- NetMon
- scalable, configurable polling system
- uses SNMP
- NodeQoS
- processes raw polled SNMP MIBdata to provide
device-level QoSindicators - NetHealth
- detects device failures and anomaloustraffic
patterns - uses on-line learning methods todetect and
classify failures - NetSyn
- synthesizes device-level QoS statisticsto
provide edge-to-edge view of QoS - NetVis
- displays QoS status of the network
- Mirage
- displays network wide QoS statistics
Applications
SEQUIN
Mirage
NetSyn
NetVis
NetMon
to network
6NetMon SNMP Based Polling
RQ size of request in bytes
RS size of response in bytes
f polling interval in secs
In the general case for dynamic
information polling bandwidth per MPLS tunnel
3.8Kbps polling Bandwidth per 10,000 tunnels
38Mbps
Therefore the choice of MIB variables to poll is
crucial
as per Bell South RFP
7NetMon Results
8NetVis MPLS Path Visualization
- Displays the physical layout of all SNMP capable
devices in the network - Selection of all LSPs between a pair of routers
using graphical interface
9NetHealth Detection of Tunnel Health
- Research Issues
- Root cause analysis for network alarms
- Alarm correlation across multiple nodes to
capture failure propagation
- Detects device and MPLS tunnel anomalies
- Notification of router overload conditions
- Detect congestion spots on interface and MPLS
tunnel - Signature based failure classification
10NetHealth Results
Visualization of traffic anomalies through color
changes
IP health
IF and MPLS tunnel health
11NodeQoS Computation of QoS Metrics
- Research Issues
- computing near real time statistical averages
based on LSP set up rate and Tunnel holding times - methods to deal with non-uniform samples due to
unreliable SNMP response
- Node level QoS information
- average load at the head end of the tunnel
- average link load for all links
- Average bandwidth usage per update period per
link or per tunnel ( in bits per sec )
T update period in sec f polling frequency in
sec
ifSpeed the speed of the link or the configured
bandwidth of the MPLS tunnel in bits per sec
12NodeQoS Results
NodeQoS Results
13Mirage Statistical Analysis of Data
- Provides interactive statistical analysis
- Built on JAVA Swing graphics library
- Suitable for studying any data set with
- numerical attributes
- multiple types / multimedia signals
- high dimensional spaces
- Examine data as isolated subsets or in context
- Data set displays
- Tables, Histograms, Scatter Plots, Parallel
Coordinates, Image Background - Cluster structure displays
- Trees, Graphs, Highlights, Colors
- Display Linkage Configuration
- All views support manual selection
- All views are linked, able to send and receive
highlight broadcast
14Mirage Online Data Analysis
MPLS Tunnels
15NetSyn Computing Edge-To-Edge QoS
In the general case
- Path level QoS information
per tunnel per link stats
N no. of tunnels using link j
t ith tunnel load on link j
ij
tunnel path info
synthesized QoS metrics
C capacity of link j
j
l total load on link j
j
- Time Synchronization of MIB information
- synchronization is achieved by using the NTP MIB.
- By running NTP, all devices on the network will
be synchronized to single clock. - This feature enables easy correlation of MIB
information obtained from multiple nodes in a
network.
16Status and Roadmap
SEQUIN ImplementationArchitecture
- Implementation environment
- Experimental Testbed
- Set up completed
- Cisco 7500 (1) Cisco 3600 (7)
- Device IOS support for both MPLS and DiffServ
- Packet generator SmartBits 2000
- Phase 1 Basic SEQUIN operation (April 02)
- GoalTo monitor and synthesize edge-to-edge
bandwidth for Cisco devices - Phase 2 SEQUIN for monitoring delay and loss
metrics (6 months) - Goal Extend the monitoring system to obtain the
class of service delays and MPLS tunnel losses
MSG Server
NetSyn
NetVis Mirage/GUI
NetHealth
NodeQoS
Working implementation
Being implemented
NetMon
17Working Relationship with NOS
- Working with Bob Murray (VitalEvent/VitalNet BU
of NOS) - Joint work on writing requirements document for
MPLS path analysis in VitalEvent Release Fall 02
(VitalEvent Systems Engineers Andrea Williams,
John Rath) - Bi-Weekly meetings (Don Hershey - VitalEvent,
Richard Haffner - VitalNet product manager) - discuss SEQUIN progress
- understand the MPLS monitoring requirements for
VitalNet poller - understand the path analysis requirements for
VitalEvent path analysis tool - Future
- Evaluate the scalability of current product and
provide input to the requirements document - Continue contributions to VN/VE to include MPLS
monitoring in their products - Help develop a scalable monitoring system for QoS
Code for MPLS path visualization
Requirements Doc
Contact with VN
Roadmap
Contact with VE
Aug01
Oct01
Jan02
Feb02
Apr02
18Proposed Enhancements to Current Vital
Architecture
- View Globally
- Reflect individual QoS requirements
- Maintain end-to-end QoS requirements
- Provide contracts
Notification Messages (CORBA, JMS)
Contract
FTP
- Aggregate Regionally
- Have certain degree of self-management
- Compute Data
- Guarantee runtime requirements
- Provide adaptation
Monitor Agents
SQL
Notification messages
Ping
Ping
Active-backup pair
- Collect Information Locally
- Provide load balancing
- Achieve fault tolerance
Pollers
19Summary
- Goal
- build a measurement-based tool for on-line
monitoring and analysis of MPLS paths in IP
networks - Research Issues
- derivation of node level QoS metrics using
existing MIB variables - synthesis of edge-to-edge QoS metrics
- scalable polling
- visualizing network wide QoS
- Status
- testbed setup working implementations of NetMon,
NetHealth, NodeQoS, Mirage and NetVis - NetSyn being implemented
20Acknowledgements
- Sanjoy Paul
- Rick Buskens
- Lawrence Cowsar
- Krishan Sabnani
- Al Aho
21Working Demo