home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Internet Info 1997 December
/
Internet_Info_CD-ROM_Walnut_Creek_December_1997.iso
/
ietf
/
bmwg
/
bmwg-minutes-95sep.txt
< prev
next >
Wrap
Text File
|
1996-06-03
|
4KB
|
151 lines
INTERIM MEETING REPORT
Minutes of an Interim Meeting on IP Provider Metrics
Benchmark Methodology Working Group (BMWG)
Pittsburgh Supercomputer Center
12 September 1995
Reported by E. Paul Love.
Guy Almes opened the meeting with a brief background of the
combination of IPPM within the BMWG. This was the preference of
the Area Directors. The co-chairs of BMWG are Guy Almes and Jim
McQuaid.
Agenda for the afternoon
Review of IPPM to date
Quick run through of the Stockholm slide presentation
Settings of priorities & assignments
Technical presentations by
Vern Paxson - LBL
Matt Mathis - PSC
IPPM to date
Pre-Danvers. Initial work in IRTF's End to End - Winter 94
through Winter 95.
BOF chaired by Matt Mathis at Danvers
Stockholm meeting has about 2 dozen participants
Mailing list now has about 300
This interim meeting
On schedule for meeting at Dallas
Stockholm slides (also in Stockholm Proceedings and via ftp at
[ftp://ftp.advanced.org/pub/IPPM/Jul-95/IPPM-Slides.ps])
Context
Growing complexity of the Internet exponential
Criteria of IP Provider Metrics
Concrete & repeatable
Useful to both users & providers
Avoid inducing artificial performance goals
No bias for equal technologies
Fair bias for non-equal technologies
Hierarchy of Metrics
Vague, gut level concept - e.g. delay
Precise, but possibly not a measurable concept
- e.g. delay of 3000 octet packet between 2 points
Precise & measurable
An Initial Set of Needed Metrics
Path Performance
Delay
Expected flow capacity for a single application
Expected aggregated flow capacity
Reliability
Delay - packet loss as they enable flow
control
Delay - jitter as it enables flow control
- multicast
Routing Stability/robustness
End to end
Stability
Quickness of recovery
Coverage
CIDR effectiveness
DNS Performance
Security of Infrastructure
NOC Responsiveness
Existing Tools
Trace route
Routing logs
Ping
ttcp
Traffic logs
Needed New Tools
Capacity tools that don't swamp production traffic
Long distance AS-path analysis tools
Better instrumentation of TCP
Relation to Emerging Infrastructure
Potential use of NAP's as locus for placing
measurement tools
Benefits
To users - when to pressure providers
To providers - knowing how to better communicate to
users, vendors & competition
Administrative Issues
IPPM is within BMWG for now
Meetings at IETF's
Between IETF's as needed
Separate WG perhaps later
Typical Agenda
Evolve consensus of metrics
Share results on tools that measure them
Share results on applications of these tools
Keep community mindful of best practices
Needed Development
General - Criteria for metrics
Pre-metric developments
Draft metric definitions
Experience ID's
An experience with tool development
An experience with tool application
[floor comment] - Do we partition routing from path
(flow & capacity) metrics?
Make sure metric definitions are same across platforms,
even if tools are different (or just similar)
High Priority Metrics
IP packet delay across paths - mean & variance
Flow capacity across paths
Single high speed flow
Aggregation of flows
Floor Comments
Who should run what?
Joe user?
Just providers on themselves?
What is the path to be measured?
Definitions of terms (rfc1242 can serve as a start)
Users don't care about packets/sec while providers may well
New things to test
Stockholm/Hawaii T3 delay-bandwidth product
Can a provider reach all portions of the global Internet?
A new tool for "Coverage."
How to publish a "Consumer's Reports for the Internet"?
Collusion from providers may be needed for routing, though not
for paths.
More metrics to consider
Packet loss
Availability of service - up time
Access - probability of a busy signal for dial in
customers
Bits/sec - raw, total vs. capacity above an average,
background load
Means of measuring need to increase a provider's
infrastructure - aggregation, muxing, etc.
Costs with performance metrics. Also, "good enough",
cheapest, etc.
Presentations
Matt Mattis - Top Down vs. Bottom Up
[ftp://ftp.advanced.org/pub/IPPM/Sep-
95/TrenoTal.ps]
Vern Paxson - Endpoint Measurement of Network Condition
[ftp://ftp.advanced.org/pub/IPPM/Sep-95/paxson-
slides.ps]