Network Performance Monitoring and Network Aware Middleware

1
Network Performance Monitoring and Network Aware
Middleware

• Warren Matthews (OIT).

2
Networking

• Campus
• Backbone group maintain 160 buildings, 1600
  switches, 52000 ports.
• Southern Crossroads gigapop (SOX)
• Provides research and commodity connectivity for
  much of the South.
• 10Gbps link to Abilene backbone connecting over
  200 research Universities nationwide.
• Southern Light Rail, Atlantic Wave.

3
Network Support

• Data Intensive Science.
• Increasing reliance on connectivity for all
  aspects of campus life.
• Impact can be significant.
• Network Performance Monitoring
• Provide the clues to solve real network problems,
  eliminate non-network issues.

4
CPR

• Campus-Wide Network Performance Monitoring and
  Recovery
• Hosts are co-located campus routers and building
  switches.
• Regular Performance Monitoring
• Set expectation.
• Multiple views of the network
• Especially the users experience.
• On demand access for backbone, operations and
  security.

5
Deployment
SOX
Ajax
Mason
Gateway Routers
EDI
Servernet
NI
ET
GLC
Rich
Core Routers
SI
845
811
DMSmith
French
Lyman
King
LAWN
OHR
OKeefe
GCATT
Skiles
FAB
SSC
6
Results - Smokeping
Average median latency is 1.8 ms
Smokeping sends 20 pings every 5 minutes
Network is slow every 15 minutes
Peak median latency is 6 ms
7
More Smokeping
Rich
EDI
SI
French
Lyman
OHR
8
Results - Nagios
Nagios monitors services such as www and email
Services do not appear to be disrupted, last
iperf achieved gt90 Mbps
Warning of high latency
9
Road Map

• Analysis and alarm system
• Plateau detector (AMP), RIPE-TT
• How should we react to alarms?
• Troubleshooting guide
• Visualization
• Front end map showing current status of all links
  mesh/tree
• More tools
• BWCTL (iperf), Traceroute, layer 3 and layer 2
• Complimented with SPAM, GOAT

10
GAMMON

• Extend CPR beyond the Georgia Tech Campus
• Web casting project
• Georgia Measurement and Monitoring
• Also a hind portion of a side of bacon or
  misleading or nonsensical talk
• SOXWatch

11
SOXWatch
SOX
Georgia Tech
Emory
ORNL
Vanderbilt
MTSU
Abilene (Atlanta)
Auburn
UA
UAB
UAH
UFL
Mississippi
Ampath
Clemson
FAU
RNP
FSU
12
Abilene (aka Internet2)
Seattle
Chicago
New York
Indianapolis
Denver
Sunnyvale
Kansas City
Washington
Los Angeles
Atlanta
Houston
13
AMI

• Abilene Measurement Infrastructure (AMI)
• Part of the Abilene Observatory
• Measurement hosts co-located with routers
• End to end Performance Initiative (e2epi)
• NLANR Active Measurement Project (AMP)

14
piPEs
Géant
MIT
ESNET
SLAC
Ohio
NOAA
Hawaii
AMI
APAN
NCREN
SOX
SDSC
RNP
15
International

• Georgia tech has adopted an international focus
  in its strategic plan
• Treat International sites just like Campus
  departments
• Metz, London, Singapore
• Across dedicated or shared links
• Networks around the world are implementing
  network monitoring
• Amsterdam

16
Dedicated Connectivity
China
St. Petersburg
RUNet
Singapore
Metz Paris
155Mbps
2.5 Gbps
SUNet
Karlskrona
Gloriad
Helsinki
1G
NorthernLight
T?Ex
Renater4
10G
NetherLight
Pacific Wave
Munich
STARLIGHT (Chicago)
Viola
GT
NLR
Barcelona Udine South Africa
Géant2
10G
2x1G
SLR
UKLight
Atlantic Wave
London
Brazil
17
Applications
Monitoring Host
Network Performance should be part of the
decision making process.
Monitoring Host
Data Store
Cluster
Resource Broker
Researcher
University
University
University
University
Shared Research Network (SoX, Abilene, Géant)
University
University
Data Store
Cluster
Monitoring Host
Monitoring Host
18
Real Applications

• Distributed Cluster
• Georgia Research Universities (GT, UGA, GSU, MCG)
• SCOOP
• Southeastern Coastal Ocean Observing Program
• SURA NMI test-bed
• ATT donated fiber
• Open Science Grid
• Measurement Information Systems WG
• CERCS?

19
Middleware

• GGF NMWG
• Global Grid Forum Network Monitoring Working
  Group
• Standardize network performance data
• Application to application
• Web Services (SOAP/XMLRPC)

20
Demo

• AMI / AMP Demo
• Advisor and Monalisa

Powered by Buzz
21
Summary

• Network Performance Monitoring
• CPR, GAMMON, SOXWatch, International
• AMI, AMP, piPEs
• SURA, Open Science Grid
• Web Services
• Global Grid Forum, Network Monitoring WG
• Analysis
• Network Aware Applications

22
This is the End.