Transcript No Slide Title

Is your Service Available?
or
Common Network Metrics
Nevil Brownlee, CAIDA
NANOG 19, Albuquerque, June 2000
Overview
• CAIDA Metrics Working Group
- Co-chairs from the networking industry
• Sue Moon (SprintLabs)
• Brett Watson (MFN/Abovenet)
• Measurement FAQ
• Service Definitions
• Common Metrics
• Availability
NANOG 19, June 2000
1 / 15
Nevil Brownlee
Metrics WG Goals
• Education
- Publish ‘Measurement FAQ’
- Publish ‘Metrics and Measurement Survey’
• Service Metrics
- Define metrics for new / emerging services
- Recommend a ‘Service Measurement Toolkit,’
encourage implementors
- Publish revised ‘Measurement Requirements
for Hardware/Software Vendors’
• User / customer participation needed !
NANOG 19, June 2000
2 / 15
Nevil Brownlee
FAQ Contents (1)
• Target audience
- Corporate users, smaller providers,
hosting service users
• Generally Accepted Terms
- Networking, types of service, faults and
failures
• Measurement Topics
- Active vs passive, one point vs many,
sampling, statistics
NANOG 19, June 2000
3 / 15
Nevil Brownlee
FAQ Contents (2)
• The Most Common metrics
- Latency, packet loss, throughput,
link utilisation, availability
• Common Measurement Tools
- ping, traceroute, SNMP, flow measurement
application monitoring, visualization
• Comparing Service Offerings
- Provider ‘net status’ pages
- Internet ‘weather maps’
- Rating services
NANOG 19, June 2000
4 / 15
Nevil Brownlee
Defining Network Service
• Service definitions (SLAs) on the Web
- Many providers publish these, e.g.
AT&T, PSInet, UUNET and MCI WorldCom
- They describe service offered to customers
• We’re only interested in describing the
service, not in contractual aspects
• Metrics used in the service descriptions
are often poorly defined
NANOG 19, June 2000
5 / 15
Nevil Brownlee
Service Example: UUNET (1)
- Network Quality
• Average monthly latency of no more than
85ms roundtrip within UUNET's network in
North America and of no more than 120ms
between New York and UUNET's international
gateway hub in London
• Comments
- Restricted to provider’s own network
- Latency not defined
- Nothing said about packet loss %
NANOG 19, June 2000
6 / 15
Nevil Brownlee
Service Example: UUNET (2)
- Service Quality
• 100 percent availability that covers the UUNET
backbone and the UUNET-ordered customer
access circuit.
• Scheduled maintenance .. will take place ..
Tuesdays and Thursdays with at least 48
hours advance notice
• Comments
- 1 minute per week = 0.01%
- Does ‘available’ time include maintenance?
- ‘100% availability’ - but what is availability?
NANOG 19, June 2000
7 / 15
Nevil Brownlee
ISP Network Report Pages
• Many providers publish these
- Sean Donelan email thread lists 13, e.g.
Abovenet, AT&T, C&W, UUNET, ELI, Jet Net ..
• Amount of detail varies
- Outage information, NOC contacts
- Latency, packet loss matrices (or averages)
• Such ‘overall’ reports don’t say much
about performance as seen from your
network
- You need some measurements at your site
NANOG 19, June 2000
8 / 15
Nevil Brownlee
Common Metrics
• Throughput, link utilisation
- Commonly measured with SNMP, RRDtool
• Latency, packet loss
- Latency is round-trip, transit + server delay
- Commonly measured with ping
• Availability
- WG definition based on IETF IP Performance
Measurement (IPPM) connectivity metrics
- Need to specify what’s available, how to
measure it, and what values are acceptable
NANOG 19, June 2000
9 / 15
Nevil Brownlee
Levels of Availability
Service availability: being able to send packets
for a specified service - say WWW request
packets - to a particular Internet host, and
to receive answering packets
Host availability: being able to send packets,
say ping packets, to a particular Internet
host, and to receive answering packets
Network availability: being able to send
packets from your network to the Internet,
and to receive answering packets
NANOG 19, June 2000
10 / 15
Nevil Brownlee
Measuring Availability
Web service availability test: download
specified pages from target web server
using web browser, measure latency,
packet loss and throughput
Host availability test: ping the target host,
having made sure that it will respond to
ICMP packets
Network availability test: traceroute to the
target host, so as to determine whether
there is connectivity to the target network
NANOG 19, June 2000
11 / 15
Nevil Brownlee
Defining Availability
• Requires negotiation between
provider and customer
- What services are important?
- What performance level is acceptable?
• Be realistic!
- Providers only control their own networks
- Some packet loss is inevitable
• Measurements are important
- Both sides should work together on this
- Make some ‘baseline’ measurements
NANOG 19, June 2000
12 / 15
Nevil Brownlee
Conclusion
• The CAIDA Metrics WG has begun by
producing its ‘Measurement FAQ’, which
provides background material on many
measurement topics
• The FAQ attempts a new definition of
Service Availability - the Metrics WG
needs feedback on this !
• The WG seeks input for its other goals,
especially for defining new metrics
NANOG 19, June 2000
13 / 15
Nevil Brownlee
More Information (CAIDA)
• Metrics WG (FAQ, mailing list)
www.caida.org/outreach/metricswg
• FAQ Contributors
Cindy Bickerstaff, Carter Bullard, Les Cottrell,
Sean Donelan, Dave O’Leary, Brett Watson, ..
• CAIDA tools taxonomy
www.caida.org/tools
NANOG 19, June 2000
14 / 15
Nevil Brownlee
More Information (Net Status)
• Provider web pages are listed in FAQ
[ISP_SERV]
[ISP_REPT]
service definitions
network performance reports
• NOC pages (search ‘xxx network status’)
www.pictureview.com/support/PVTS2.html
www.psinet.com/netstatus/
www.sprintlink.net/netstat.html
• Network performance pages
stats.sjc.above.net/traffic/
ipnetwork.bgtmo.ip.att.net
traffic.cwusa.com/index.html
NANOG 19, June 2000
15 / 15
Nevil Brownlee