100 likes | 217 Views
OAM – an end-user perspective. Presented by: Yaakov (J) Stein RAD Data Communications Ltd. Unique Access Solutions. FM and PM. Traditionally, a distinction has been made between 2 OAM functionalities F ault M onitoring (FM) OAM runs continuously/periodically at required rate
E N D
OAM – an end-user perspective Presented by: Yaakov (J) Stein RAD Data Communications Ltd. Unique Access Solutions
FM and PM Traditionally, a distinction has been made between 2 OAM functionalities • Fault Monitoring (FM) • OAM runs continuously/periodically at required rate • detection and reporting of anomalies, defects, and failures • used to trigger mechanisms in the • control plane (e.g. protection switching) and • management plane (alarms) • required for maintenance of basic connectivity • Performance Monitoring (PM) • OAM before enabling a service or when service quality degrades • measurement of performance criteria (delay, PDV, etc.) • required for maintenance of Quality of user Experience (QoE) In traditional Connection Oriented (CO) networks these functionalities are relatively non-overlapping although Packet Loss Ratio (PLR) is in both : • low ratios in PM • high ratios in FM
From CO to CL Networks The ITU defines a connection in a CO network as an association between an input and output ports (topological definition – agnostic to distance between input and output ports) A trail (ITU-T definition) is basically • a connection with OAM For ConnectionLess (CL) networks the ITU-T defines a flow as a group of packets with common routing (probable interpretation: packets • coming from the same input • going to the same output • that require the same treatment <including routing> Isn’t this a FEC ?) A connectionless trail is • a flow with OAM But do these CL definitions make any sense ?
Meaningful flows For the concept of a flow to be meaningful in this context There need to be enough packets in each flow • one or two packets do not constitute a meaningful flow ! Packets in each flow need to be sufficiently related • otherwise it does not make sense to treat similarly (QoE) • difficulties - new waists of Internet hourglass Flows need to be recognizable by the Network Elements (NEs) • NEs that insert/monitor the OAM need to recognize the flows • e.g., an IP flow could be defined by 5-tuple and time • difficulties : • peer-to-peer file transfer – many 5-tuples for single application • L2/3-VPNs – many users may use the same 5-tuples • encryption may hide port numbers
Flow lengths – empirical evidence To test the number of packets in a typical flow We recorded headers from an international link of an ISP We sampled • both residential and business IPVPN • at various hours of the day We analyzed the number of packets in a “unidirectional flow” • common 5-tuple (deeper DPI did not substantially change results) • no lapse for 5 seconds (Gb interface carries about 1.5M pps) Analysis still a work in progress
Residential Internet – flow length distribution (1) • most flows VERY short (>50% single packet) • average length 7 • long tail (power law) P(L) L-1.85 • some length flows (UDP and peer-to-peer) • Conclusion only small fraction of packets are in flows 5 sec of traffic on Gb link
Residential Internet – flow length distribution (2) • more BW in VERY short than longer ( 3% BW are single packet) • average length 400 • for long lengths BW flow length • Conclusion sizeable small fraction of BW are not in flows 5 sec of traffic on Gb link
IPVPN – flow length distribution • most flows VERY short (>50% single packet) • average length (weighted on flows, not BW) 40 • long tail (power law) • usually no length flows (and practically no peer-to-peer) • Conclusion Most packets are in very short flows (too short to warrant classification)
Proposals (1) • leave most FM to lower levels - Ethernet, MPLS-TP, SDH/OTN, but not IP where it is needed for : • FRR/protection • SLAs enforcement • etc. and where the needed functionality is mostly already defined • end-users need a rather limited set of tools : • FM for the access network • PM for the access network and the end-end path don’t over-engineer • IPPM has developed methodologies and tools don’t re-invent the required measurements can be performed by TWAMP-lite reflectors in (or very close to) edge routers
Proposals (2) • OAM needs to determine/measure : • for access network only : • [FM] connectivity (are we getting out to the Internet ?) 1 CC per second is usually sufficient Note: this can be indicated to user as limited connectivity and statistics collected for SLA enforcement • for both access leg and end-end path • [PM] unidirectional data rates • [PM] round-trip latency access network