1、 ATIS-0100016 END-TO-END SERVICE AVAILABILITY: GENERAL DEFINITION TECHNICAL REPORT The Alliance for Telecommunication Industry Solutions (ATIS) is a technical planning and standards development organization that is committed to rapidly developing and promoting technical and operations standards for
2、the communications and related information technologies industry worldwide using a pragmatic, flexible and open approach. Over 1,100 participants from over 300 communications companies are active in ATIS 22 industry committees and its Incubator Solutions Program. Notice of Disclaimer February 2008,
3、work in progress). 5 GENERAL DEFINITION OF END-TO-END SERVICE AVAILABILITY A communications network is an entity that connects users who access the network through ingress and egress points. The network is composed of various network elements and links that connect them. However, to define end-to-en
4、d service availability at the highest conceptual level, the network can be treated as a single entity with ingress and egress points that allow access to the network by its users. Service availability is the fraction of time that a service is considered available as supported by a system. In the cas
5、e of end-to-end service availability for a network, the system is a communications network and the definition must be extended to represent all access points to the network. End-to-end service availability for a communications network is the fraction of time that service is available between an arbi
6、trarily-specified ingress point and arbitrarily-specified egress point to the communications network. Typically, an access point is expected to be an edge network element; 1This document is available from the Alliance for Telecommunications Industry Solutions (ATIS), 1200 G Street N.W., Suite 500, W
7、ashington, DC 20005. 2This document is available from the International Telecommunications Union. ATIS-0100016 however, the level of granularity can be adjusted as needed. For example, an access point could be an access port if the calculation or data demands that level of granularity. Given a netwo
8、rk with a set of access (ingress and egress) points, a specified service, and a specified time window, the definition of end-to-end service availability for the network supporting that service over the duration of the time window is: End-to-End Service Availability =NiNjijNiNjijijwfw1111where fijis
9、the fraction of time that service is available between ingress i and egress j (see 4.3) and N is the number of access points to the network. Note that the distinction between fijand fjiallows for the possible difference in originating and terminating service. The wijweights are needed when the acces
10、s points are different sizes (e.g., different numbers of users, different communication bandwidths) and/or have been in-service for different periods of time. The inclusion of separate wijand wjiweights allows for differences in the direction of service; for example, if service goes from i to j but
11、not j to i, then wij wji= 0. The following sections describe the key elements of the definition: 1. Specifying the communications network and its access points. 2. Specifying the wijweights for access point pairs. 3. Specifying service availability fijbetween access points. 5.1 The Communications Ne
12、twork In order to define end-to-end service availability, the network must be clearly defined. The limits of the network (i.e., demarcation points) as defined by the access points must be clearly understood. It is recognized that services over IP networks (e.g., VoIP services) may span several provi
13、der network domains. In such cases, the assignment of availability for individual providers becomes an issue that must be dealt with. It is proposed that initial work should focus on assessing availability over a single providers network, and that later efforts can concentrate on tying together end-
14、to-end availability over several network domains. The definition of a service outage needs to be agreed upon. Specifically, when is a service considered to be unavailable? This depends on the type of service. In general, it is simplest and most practical to define outage as a binary phenomenon - i.e
15、., the service is either “up” or “down” based on a specific defined threshold. However, it may be desirable, in a particular evaluation, to consider the possibility of a “partial outage” e.g., an outage that permits limited service delivery due to inadequate bandwidth or network resources. In such c
16、ases, agreement needs to be reached on how to define the fijfractions in the availability definition above. These fractions could be “pro-rated” according to the degree of partial availability, if reasonable estimates of such pro-rations are available. Alternately, the fractions could be assigned bi
17、nary values: 3 ATIS-0100016 4 1, if X% of traffic is estimated to be conveyed between the end points where the value X = 50, for example. 0 otherwise. The definition of service outage and partial outage for each service is beyond the scope of this document. However, any calculation of end-to-end ser
18、vice availability should include a description of the definition of service outage used, as well as how partial outage was estimated if the concept was used. 5.2 Weights for Access Point Pairs Access point weights wijshould be defined based on the sizes of the access points and the time that both ac
19、cess points were in service. One possible method for defining the weight for service between access points i and j is: wij= si sj Dijwhere si(sj) is the size of access point i (j) and Dijis the duration of time the network has supported access points i and j in the time period TS, TE. For example, c
20、onsider a wireline circuit network for voice service where the time window of interest is the year 2004 (i.e.,. TS, TE = January 1, 2004, December 31, 2004). Two access points could be local switches A and B with 30,000 access lines for A and 20,000 access lines for B. Switch A was in-service for al
21、l of 2004 while Switch B was in service for the last 100 days of 2004. In calculating the end-to-end service availability for voice service in this network in 2004, the weight for service availability between switches A and B is: wAB= 30,000 x 20,000 x 100 = 60,000,000,000. Access point size may dep
22、end on the type of access element, type of network, and/or type of service. This weight definition requires that access points use the same unit dimensions in defining size. This may require conversions for access points with different unit dimensions. Other definitions for weighting can be consider
23、ed in the discussions. For instance, in IP networks, port bandwidths can be added together to represent the weight between the two port end points. 5.3 Service Availability Between Two Access Points The service availability fijis the fraction of time that service was available between the ingress po
24、int i and the egress point j; the ingress point i is the point at which service was initiated and the egress point j is the point to which ingress point i desires communication. This fraction encompasses any impacts on service, including impacts from loss of point-to-point connectivity or any failur
25、e in the network that affects the users ability to communicate. Given the time period of interest was between times TSand TEand given M periods of service outage from ingress point i to egress point j were identified with durations O1, O2, O3,OM, then the service availability fijis : ATIS-0100016 fi
26、j= 1 for M = 0 or ijMmmijDOf=11 for M 0 The definition of service outage period between two access points varies from service to service. These definitions will be addressed separately in each service-specific document. In general, a service outage includes inability to initiate a service attempt, i
27、nability to terminate a service attempt, unsatisfactory quality for service in progress, and interruption of service in progress. By varying the definition of service outage, end-to-end service availability at different grades of service can be estimated. The definitions should consider failure mode
28、s to be addressed; the metric could encompass any service outage, including those resulting from hardware failures, software failures, physical damage (e.g., fiber cuts), procedural errors, database errors, maintenance, and traffic congestion. 5.4 Example Calculation of End-to-End Service Availabili
29、ty Consider a small network of N=5 access points with the following sizes in terms of thousands of lines: Access Node i Size si1 10 2 20 3 30 4 40 5 50 We wish to calculate the end-to-end service availability for this network over the past year. All access nodes except Access Node 1 were in-service
30、for the full year; Access Node 1 was in-service only for the last six months. The in-service time Dijfor all access point combinations is shown in the following table in months: Dij(months) Access Node 1 Access Node 2 Access Node 3 Access Node 4 Access Node 5 Access Node 1 6 6 6 6 6 Access Node 2 6
31、12 12 12 12Access Node 3 6 12 12 12 12 Access Node 4 6 12 12 12 12Access Node 5 6 12 12 12 12 5 ATIS-0100016 The weights wij= sisjDijfor the calculation are: wijAccess Node 1 Access Node 2 Access Node 3 Access Node 4 Access Node 5 Access Node 1 600 1200 1800 2400 3000 Access Node 2 1200 4800 7200 96
32、00 12000 Access Node 3 1800 7200 10800 14400 18000 Access Node 4 2400 9600 14400 19200 24000 Access Node 5 3000 12000 18000 24000 30000 The sum of the weights in the table is =NiNjijw11= 252600. The availability fijof service between access point pairs, using empirical data as measured over the year
33、 under consideration, is shown in the following table: fijAccess Node 1 Access Node 2 Access Node 3 Access Node 4 Access Node 5 Access Node 1 0.99999 0.99990 0.99995 0.99999 0.99999 Access Node 2 0.99991 0.99999 0.99999 0.99999 0.99999 Access Node 3 0.99995 0.99999 0.99999 0.99998 0.99999 Access Nod
34、e 4 0.99999 0.99999 0.99998 0.99999 0.99997 Access Node 5 0.99999 0.99999 0.99999 0.99997 0.99999 The following table shows the products of availability and weight for each access point pair (wijfij): wijfijAccess Node 1 Access Node 2 Access Node 3 Access Node 4 Access Node 5 Access Node 1 599.994 1
35、199.880 1799.910 2399.976 2999.970 Access Node 2 1199.892 4799.952 7199.928 9599.904 11999.880 Access Node 3 1799.910 7199.928 10799.892 14399.712 17999.820 Access Node 4 2399.976 9599.904 14399.712 19199.808 23999.280 Access Node 5 2999.970 11999.880 17999.820 23999.280 29999.700 The sum of the val
36、ues in this table is =NiNjijijfw11= 252595.878. 6 ATIS-0100016 The end-to-end service availability for this network over the year considered is: =NiNjijNiNjijijwfw1111= 878252595252600.= 0.99998368. 6 EQUIVALENT DEFINITION OF END-TO-END SERVICE AVAILABILITY 6.1 Definition in Terms of Weighted Downti
37、me The definition of end-to-end service availability is given in terms of a summation over access point pairs. This is useful conceptually as it defines the end-to-end service availability for the network as a weighted average of the service availability of access point pairs. At times, it can be us
38、eful to apply the definition in terms of weighted downtime: End-to-End Service Availability = =NiNjijNiNjijijwfw1111= =NiNjijMmmmwOS1111 where M service outages with durations O1, O2, O3,OM, occurred in the network during the time period TS, TE. Each outage duration has a corresponding weight Smrefl
39、ecting the number of access point pairs affected by the outage and the sizes of the access points affected: =NiNjjiijmmssIS11where Iijm= 1 if service from access point i to access point j was unavailable from outage m and 0 otherwise. 6.2 Example of Prediction of End-to-End Service Availability The
40、above definition of end-to-end service availability as the weighted sum of outage durations is useful as the basis for prediction. Consider a communications network with N = 50,000 access points each serving 1,000 users. The network also consists of 200 distribution network elements which provide th
41、e access points with an interface to a core backbone IP network. Modeling of the network predicts an average of 3302.2 service outages per year as a result of failures in the network. The table below describes the service outages resulting from these failures: 7 ATIS-0100016 8 Failure Outages Durati
42、on (mins) Access Point Pairs Affected Product Access Point 3000 90 99999 26999730000 Distribution Facility 300 300 99999 8999910000 Distribution Element 2 180 24937500 8977500000 Core 0.2 180 250000000 9000000000 SUM= 53977140000 The first line indicates that modeling predicts an average of 3000 ser
43、vice outages per year, each having a duration of 90 minutes on average; each outage affects 99,999 access point pairs (50,000 pairs originating from the failed access point and 49,999 terminating to the access point, thus eliminating double counting of the failed access point). The other lines of th
44、e table are calculated in a similar fashion. The final column is the product of the other three columns. Dividing the sum of this column by the number of access point pair minutes in a year: Access point pair minutes in a year = 50,0002x 365 x 24 x 60 = 1314000000000000 and subtracting the result fr
45、om 1 gives a predicted end-to-end service availability of 99.9959%. NOTE - To simplify the calculation of the example, the access point size of 1,000 users has been ignored; this can be done here since all access points are assumed to have the same size. Assuming Poisson distributions for the number
46、s of outages per year and exponential distributions for outage durations, a distribution of end-to-end service availability for this network over a year of service can be estimated using Monte Carlo simulation. The table below shows key quantiles for the distribution of availability in the example:
47、Quantile .005 .025 .05 .5 .95 .975 .995 Availability (%) 99.9806 99.9891 99.9915 99.9966 99.9973 99.9973 99.9974 The distribution shows that in about half of the years, end-to-end service availability is slightly better than the average of 99.9959%. However, it also shows that a significant chance e
48、xists for this availability to be much lower than the average. The skewed nature of this distribution should be considered when setting SLAs for the network. 7 ESTIMATION OF END-TO-END SERVICE AVAILABILITY Exact measurement of end-to-end service availability for a network over a specified time perio
49、d would require continual monitoring of the service capability for each pair of access points in the network. In ATIS-0100016 9 most cases, such a measurement process would be a prohibitively difficult endeavor. In such cases, sufficient results can be achieved through sampling service capability over the user base and time. It may be possible to estimate end-to-end service availability based on user experience (e.g., user calls blocked, packets lost from user attempts), but such results may require statistical adjus
