Packet Loss measurement Model

Today, Performance monitoring or tracking of the performance experienced by customer traffic is a key technology to strengthen service offering based on enhanced QoE and SLAs. The lack of adequate tools to detect an interesting subset of a Packet Stream, as identified by a particular packet attribute(e.g., commit rate or DSCP) and measure that packet loss drives an effort to design a new method for the performance monitoring of live traffic, possibly easy to implement and deploy. The draft aims to define loss measurement matrix models for multiple customer service flows on the network. Each customer service flow is corresponding to an interesting subset of the same packet stream. The customer or packet stream can be identified by a list of source or destination prefixes, or by ingress or egress interfaces. The network would be provisioned with multiple services(e.g., real time service, interactive service) having different SLAs(e.g., bandwidth constraint or end packet loss constraint for the end to end path) based on the customers' requirement. This models aims at computing Loss measurement for these services (belonging to the same customer)independently for each defined SLA matrixes. The class-of-service and packet color classification defined in the network is a key factor to classify network traffic and drive traffic management mechanism to achieve corresponding SLA for each service. This draft uses the class-of-service model and color based model for any given network to define the packet loss measurement for various services with the different SLA requirements. The proposed matrix models is suitable mainly for passive performance measurements but can be considered for active and hybrid performance measurements as well. This solution models loss measurement in different kinds of network scenarios. The different models explained here will help to analyse packet loss pattern, analyze the network congestion in a better way and model the network in a better way. Loss measurement is carried out between 2 end points.The underlying technology could be an active loss measurement or a Passive loss measurement. Any loss measurement will require 2 counters: Number of packets transmitted from one end point. Number of packets received at the other end point. This draft explains the different ways to model the above data and get meaningful result for the loss measurement compulation. The underlying technology could be an MPLS Loss measurement, or based loss measurement or an IP based loss measurement.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC2119. An Observation Point is a location in the network where data packets can be observed. Examples include a line to which a probe is attached, a shared medium, such as an Ethernet-based LAN, a single port of a router, or a set of interfaces (physical or logical) of a router. The persistence Data store is a scalable data store which collects time based data such as streaming data or time series data for network analytics. Time Series Data is a sequence of data points with time stamps. The data points are limited to loss measurement results in this document. A Packet Stream denotes a set of packets from the Observed Packet Stream that flows past some specified point within the Metering Process. An example of a Packet Stream is the output of the Selection Process. The Packet Content denotes the union of the packet header (which includes link layer, network layer, and other encapsulation headers) and the packet payload. It is used to identify the color that applies to the data packet. Color identifier can be assigned to service level packet based on commit rate and excess rate set for the traffic. For example, the service level packet will be set with "green" color if it is less than committed" rate; the Service Level packet will be set with "yellow" color if it is exceeding the"committed" rate but less than the "excess" rate. The service frame will be set with "red" color if it is exceeding both the "committed" and "excess" rates. It is used to identify the COS that applies to the data packet.CoS identifier can be assigned based on dot1p value in C-tag, or DSCP in IP header. Complete data measurement is a data measurement method which monitors every packet and condense a large amount of information about packet arrivals into a small number of statistics. The aim of "monitoring every packet" is to ensure that the information reported is not dependent on the application. Color based data measurement is a data measurement method which monitors the data packet with the same color identifier. COS identifier could be C-Tag Priority Code Point(PCP) or DSCP. COS and color based Data measurement is a data measurement method which monitors the data packet with the same defined SLA matrix.The SLA matrix is an array of Color identifier attribute and COS identifier attribute.

A stream of packets is observed at an Observation Point of the source endpoint and destination endpoints. Two observation points can also be placed at the same endpoint for node monitoring , i.e.,one is at ingress interface of the endpoint and the other is at the egress interface of the endpoint. A Selection Process inspects each packet to determine whether or not it is to be selected for data analytics. The Selection Process is part of the Metering Process, which constructs a report stream on selected packets as output, using the Packet Content, and possibly other information such as the arrival timestamp. The report stream on selected packets will be stored in the persistence data store for real time data analysis or time sequence data analysis. The following figure indicates the sequence of the three processes (Selection, Metering, and Storing).

+-----------+ +-----------+ |Persistence| |Persistence| |Data Store | |Data Store | Src Endpoint +-----^-----+ Dst Endpoint +------^----+ +------------------+ | +------------------+| | Metering Process | | | Metering Process || Observed | +-----------+ | | | +-----------+ || Packet--->| | Selection |------+ Observed | | Selection | || Stream | | Process |--------Packet--->| | Process |-----+ | +-----------+ | Stream | +-----------+ | +------------------+ +------------------+

This section defines the Selection Process and related objects. A Selection Process takes the Observed Packet Stream as its input and selects a subset of that stream as its output. A Selection Process may maintain state information for use by the Selection Process. At a given time, the Selection State may depend on packets observed at and before that time, and other variables. Examples include sequence numbers of packets at the input of Selectors,a timestamp of observation of the packet at the Observation Point,indicators of whether the packet was selected by a given Selector. A Selector defines the action of a Selection Process on a single packet of its input. If selected, the packet becomes an element of the output Packet Stream. The Selector can make use of the following information in determining whether a packet is selected: COS Identifier in the Packet Content; Traffic attribute such as Color identifier; Combination of CoS Identifier and Color Identifier

A Metering Process selects packets from the Observed Packet Stream using a Selection Process, and produces as output a Report Stream concerning the selected packets.

This model uses the complete data traffic between the 2 end-points to compute loss measurement. This will result in computation of loss measurement for the entire traffic in the network in one direction. This is primarily used in cases of backbone traffic where traffic from different services are aggregated and send into the core network.This will count all the packet, this gives the overall loss measurment between one endpoint to other.

This is same as the above section of "complete data measurement" with a minor difference, only monitoring the data packet with specific color identifier. In this model the packets are counted in the following Way: Count specific data traffic with different color identifier between 2 end points for loss measurement.One example of Color based data measurement is to count two type of color based traffic: Count all committed traffic between the 2 end-point for loss measurement. Count all Excess traffic which is beyond the committed traffic for the specific network. When both of these are combined then it becomes the model for complete traffic as mentioned in the above section. In practice the Color of traffic can be using any mechanism based on the network encapsulation.As long as the packets could be treated differently based on the underlying encapsulation this mechanism could be used. This is used in core networks where the aggregated traffic has differential priority and loss measurement can be computed on the committed traffic which is guaranteed in the network when compared with excess traffic which could be dropped based on network load and provisioning.

This model uses the data traffic in the network which is flowing in a specific COS to measure the loss in the network.Based on the class of traffic in the network the transmitted and received packets are counted to calculate the loss measurement. Primary use of this kind of loss measurement is to measure loss measurement for a specific service which has strict SLAs. The service could be a point-to-point layer2 service, an MPLS based service.

This model uses a combination of both Color based data measurement and Cos based data measurement. Packets are counter for a specific COS with a specific color.This can count both in profile packet which are green and yellow which are out profile packets. This will not count the red packet which violates the SLA.This will count the packet for each SLA and color separately.

This model reinforces the use of well known methodologies for passive performance measurements.A very simple, flexible and straightforward mechanism is presented in . The basic idea is to virtually split traffic flows into consecutive batches of packets:each block represents a measurable entity unambiguously recognizable thanks to the alternate marking. This approach, called Alternate Marking method, is efficient both for passive performance monitoring and for active performance monitoring.

Consider a provider running point to point service between router A and B for his customer "X".Customer "X" has voice traffic which requires special treatment,then he requires attention for database traffic. The customer "X" has SLA with the provider. Now the challenge faced by the provider is how to measure the traffic of customer "X" for each class and calculate the bandwidth, moreover the provider has to see whether the "X" is sending traffic which is exceeding the level so that he can make tariff accordingly. This problem is solved by the above models which can measures the packet for each class of traffic and tabulates the data. Later point of time this data can be pulled for evaluation.

+-------+ +-------+ | | | | | +--------------+ | | | P2P service | | +-------+ +-------+ Router A Router B The same considerations can be applicable in a multipoint to multipoint scenario (e.g. VPN or Data Center interconnections). In this case Customer "X" has multiple ingress endpoints and multiple egress endpoints. The proposed matrix model is composed by the number of flows of "X" in the multipoint scenario and by class-of-service and color classification. So the SLA matrix is a reference for the analysis and evaluation phase.

We would like to thank Brian Trammell for giving us the opportunity to present our draft.We would like to thank Greg Mirsky for the comments.

This document does not introduce security issues beyond those discussed in [I.D-ietf-idr-ls-distribution] and [RFC4271].

IANA maintains the registry for the TLVs. BGP TE Performance TLV will require one new type code per TLV defined in this document.