Transcript Performance Monitoring Infrastructure

Performance monitoring
TF-NGN meeting,
Rome (Italy), 06-02-2003
Nicolas Simar, Network Engineer
DANTE
Performance monitoring -- Nicolas Simar ([email protected])
Logical Model
Domain central tool
measurements
get data
Inter domain
communication
Performance monitoring -- Nicolas Simar ([email protected])
Logical Model
Communication protocol for data exchange
(IPPM, e2epi)
Implication on measurement
collection presentation?
Measurement protocol
standardisation (IPPM)
measurements
get data
Inter domain
communication
Performance monitoring -- Nicolas Simar ([email protected])
Inter domain communication
Data exchange
• Need to have a standardise way of exchanging data
between domain
– see what IPPM is doing
– see what e2epi is doing
– anybody/anything else?
• Requests and message format should be understood
by every domain.
Performance monitoring -- Nicolas Simar ([email protected])
Inter domain communication
Inter domain measurement
• Need to have a standardise way of measuring the
metrics, measurement point compatibility
– see what IPPM is doing
– anybody/anything else?
• Need to have a defined packet structure.
Performance monitoring -- Nicolas Simar ([email protected])
Metric definition
Singleton - Accuracy - cfr website
• OWD - 1ms
• OWPL - O(10^-(x+1), 10^-(x-1))
• IPDV - 1ms
• Re-ordering
– Two definitions at IETF
– Interresting to histogram per “level of re-ordering”
• traceroute
• RTT ? - 1ms
– for backward compatibility with existing systems
Performance monitoring -- Nicolas Simar ([email protected])
Metric definition
Services monitored
• Measurement should be done between two points
for ... (depending on service availability - up to each
domain)
– IPv4 and IPv6
– BE, Premium and LBE
Performance monitoring -- Nicolas Simar ([email protected])
What do can be presented to who?
• End user
– current metric value
– metric history
• Planning
– same as end-user
– should be able to access raw data
• NOC
–
–
–
–
same as planning
should be able to see the packets incoming in real time
should be able to monitor a particular session
should be able to set up a new session
Performance monitoring -- Nicolas Simar ([email protected])
Statistics
• Exchange statistics and/or raw data between
domains
• Which statistics do we want to provide (which ones
are the most useful?)
–
–
–
–
–
Average, SDR (does it work on any sampling methods)
Average max, average, average min
Absolute max, min
Over which period?
For one given packet size in continuous mode
Performance monitoring -- Nicolas Simar ([email protected])
Statistics
• Statistic showed and exchanged between domains
on which time scale (1min - 5 min)?
– Statistics additivity/concatenation across several domains
– Statistics accuracy across several domains
Sampling
• How do we want to sample?
– Is there anybody which has already study it?
Performance monitoring -- Nicolas Simar ([email protected])
Domain central tool
Functions descriptions - 1st draft
• Communication module
• Request “analysor” module
• Path finder module
• Data “recuperator” module
Performance monitoring -- Nicolas Simar ([email protected])
Domain central tool
Functions descriptions - 1st draft
• Communication module
– with other domains and domain web-interface
– check message format validity, authentication, request info
to “request analysor” and send the result
Performance monitoring -- Nicolas Simar ([email protected])
Domain central tool
Functions descriptions - 1st draft
• Request “analysor” module
– receive a information request from communication module
– check if this data is available in the domain
– if yes, ask to the “path finder” module from which Point of
measurement (PM) the data should come from.
– ask to the “data recuperator” to retrieve data (either in a
centralised DB, in a distributed DB or on the PM itself)
– send information requested to communication module.
Performance monitoring -- Nicolas Simar ([email protected])
Domain central tool
Functions descriptions - 1st draft
• Path finder module
– receive a information request from communication module
– check according to the information received from the request
“analysor” module which PMs and which domain ingressegress are involved on the path and send back this
information
– module should be linked to a network representation either
via a database or to some routing information
– some parts of this module are model specific
• Quite challenging part :-)
Performance monitoring -- Nicolas Simar ([email protected])
Domain central tool
Functions descriptions - 1st draft
• Data “recuperator” module
– consists of several module to retrieve the data, up to one per
type of information having to be retrieved (these modules
are domain specific)
Performance monitoring -- Nicolas Simar ([email protected])
Date storage in a domain
• Raw data distributed in PM
• Central database (to avoid too frequent access to
MPs)
– Statistics (statistics over which period of time, which
statistics)
– raw data (database size)
Performance monitoring -- Nicolas Simar ([email protected])
Measurement box wish list
• Wish list
– by mid March
– any contribution/advices/help welcome
– metric, parameters, sampling, security, data storage, data
accessibility, future evolution, accessibility, price, additional
measurements, real time access of packet, 2 NIC (one for
measurement, one to access data), scalability, else?
Performance monitoring -- Nicolas Simar ([email protected])
Concept test
• Need of at least three consecutive domains having
partially implemented MPs (GARR, GÉANT,
HEANET, SWITCH)
– concept validation
– software being build on top of it (DANTE has some
resources)
Performance monitoring -- Nicolas Simar ([email protected])
Miscelaneous
• new mailing list with history feature (thanks to
SWITCH :-)
• Contact e2epi to see if both systems can be bridged,
Intermon project.
• Investigate which NRENs are interested by the
concept to get their input
• Investigation of graphical representation
• Web page: http://www.dante.net/tf-ngn/perfmonit
Performance monitoring -- Nicolas Simar ([email protected])