Performance monitoring – Rome meeting minutes

The metric singleton definition are defined on http://www.dante.net/tf-ngn/perfmonit/metrics.html

• one-way packet loss:
  VR: target around 10^-4 to represent what the application could need (which applications?)
  SL: need to send 10^5 packets to reach this accuracy (so that kind of statistic can only be calculated over a long period of time).
  VR: not suitable for high speed TCP which are looking for packet loss around 10^-12
  
• packet re-ordering
  SL: two definition at the IPPM. Shalunov easier to implement.
  NS: need to have some histogram indicating the number of packets re-ordered per level of re-ordering (x packets reordered from one position, z from two position, etc)
  No consensus, somebody would need to study it and report to IPPM what we would prefer.
  
• OWD accuracy = 1ms
• IPDV accuracy = 1ms
  SL: must be measured between two packet of the same size
  VR: would like to do some measurement for small, medium and large packet size (as well as jumbo). What does it implies (?)
  SL: for most metric packet size will not matter (?)
• RTT needed for backward compatibility with existing systems deployed in some NRENs.
• Traceroute in parallel to the measurements must be done.

Metric representation and exchange between domains

NS: average over the last x min (one minute would be nice) and history, three types of user: end-user, planning and NOC.
VR: history over different scale would be nice.
SL: exchange raw data between domains (because don’t know which is the best way to represent them) as well as statistics (eg beacon)
MC: history and amount of data in database based on storage capacity
  Keep raw data for at least two week for debugging.
  Should reflect the SLA to the end user (eg if value provided every 5min or every minutes)
  Compare avg against different average over longer period.
SL: create a web service to exchange data
VR: like NIMI implementation (see his presentation in Amtsterdam)

NS: presentation of a domain "central" tool.