Oftel Discussion Paper:

As promised in the original Direction for FRIACO issued on 26 May 2000, Oftel proposes to conduct a review of the interconnection charge with particular attention given to the methodology used for the Adjustment Ratio.

A meeting will be held at Oftel at 3:00pm on Monday 20 November 2000 to go through and discuss the proposals with Other Licenced Operators (OLOs) and ISPs. In order that we can make the necessary accommodation arrangements please advise Aimee Moore, here at Oftel, of the names of attendees from your organisation, either by telephone on: 020 7634 8947 or by e-mail  before 3:00pm on Friday 3 November.

1 Introduction

1.1  Flat Rate Internet Access Call Origination (FRIACO) is a product which provides virtual capacity from the concentrator up to a port on the network side of the digital local exchange (DLE), referred to as a FRIACO port. Unlike metered (pence per minute) call origination, the FRIACO charge is a fixed fee per unit of capacity in the FRIACO port and does not vary with the volume of traffic carried on this capacity. The charge per FRIACO port of 64 kbit/s determined in Oftel’s Direction of May 2000 was £424.25.

1.2  The Direction included a commitment to undertake a review of the FRIACO charge, drawing on the experience of the operation of FRIACO:

    "7. The Charge shall be reviewed by the Director by reference to information on Internet Traffic using FRIACO in the
    period to 1 December 2000, unless the parties otherwise agree in writing."

1.3  In outline, the FRIACO charge was derived by Oftel in two stages:

     a  The average cost of a call origination circuit (from concentrator to FRIACO port).
     b  An adjustment ratio, multiplied by (a), reflecting the number of call origination circuits required per FRIACO port.

1.4  The purpose of this discussion paper is to raise methodological issues about the calculation of the adjustment ratio and to invite comments on how the updating of the adjustment ratio might best be addressed in the Review. Comments about the data on traffic volumes and patterns that Oftel should seek to obtain in the Review would also be welcome.

1.5  The outline of the discussion paper is as follows. The next section discusses why an adjustment ratio is needed and explains the adjustment ratio that was used in the FRIACO determination in May 2000. Section III discusses some measurement issues, including the ‘multiple Busy Hour effect’ that is especially relevant to LECO circuits. In the subsequent section, using a series of illustrative numerical examples, the formula for the adjustment ratio is compared against the correct adjustment ratio, as the assumed profile of FRIACO traffic is allowed to vary. The conclusion of the paper suggests that the theoretically correct adjustment ratio is complex to derive and depends upon the pattern of FRIACO traffic across times of day and LECO circuits.

2 Adjustment Ratio used in the FRIACO Determination

2.1  The purpose of the adjustment ratio is to measure the number of virtual Local Exchange Call Origination (LECO) circuits, i.e. between concentrator and DLE, which are required per FRIACO port at the DLE. It is expected that an OLO will run a FRIACO port at a higher utilisation than is achieved by BT on LECO circuits (for reasons including economies of scale, higher Grade of service requirements on PSTN traffic and multiple busy hours – see the Explanatory Document to the Direction for further details). The utilisation rate of circuits is measured by Erlangs per circuit (EPC). Lower EPC for LECO circuits relative to EPC for interconnect ports means that in any given hour more than one LECO circuit is being used per port. The number of LECO circuits needed will be equivalent to the EPC per FRIACO port divided by the EPC per LECO circuit.

2.2  The adjustment ratio used in the Determination was the ratio of (expected) EPC on FRIACO ports to the EPC measured on LECO circuits. Table 1 below shows how the EPCs used in the adjustment ratio in the Determination were derived. The average EPC for LECO links in the network BH was measured by BT as 0.36. The average EPC for internet traffic over OLOs’ interconnect ports was measured by BT as 0.78. This was measured in the individual BHs of the interconnect ports. Oftel assumed that this utilisation rate also applied in the network BH. The EPC of interconnect ports in the network BH was not directly measured. It is feasible that the EPC of interconnect ports in the network BH could have been lower than 0.78 (ie the timing of the interconnect port BHs could have been non-coincident with the timing of the network BH). This would have resulted in a lower price for FRIACO at the DLE.

Table 1: EPCs in the Determination

  Note: EPCs measured on basis that there are 250 BHs in the year.

2.3  There are a number of reasons for the assumption that the BH for the interconnect ports and the PSTN network coincided. Firstly, Oftel did not feel that the assumption was unrealistic. PSTN traffic on BT’s network has three peaks – roughly corresponding to peaks in the morning, afternoon and evening – so there was some probability that the FRIACO peaks and PSTN peaks might coincide. Even if FRIACO did not exactly coincide with the PSTN peaks, there was a possibility that FRIACO traffic would hit the shoulder of one of these peaks. If FRIACO turned out to be successful and account for significant traffic volumes, then FRIACO would be likely to convert the shoulder of a peak into a new PSTN peak.

2.4  Secondly, Oftel believed that using a low initial value could have led to undesirable consequences. An initially low value would have led to instability in the price of FRIACO over time. This was counter to the aim of establishing a fairly stable flat rate charge. Furthermore, the lower initial price would have led to greater uncertainty about future prices which could distort OLOs investment decisions. Moreover, if OLOs were to extend their networks and plan their business cases on the basis of this low price, any upward adjustment in the future price may have had an adverse effect on their business plans and the sustainability of their businesses.

2.5  Oftel also made one other implicit assumption in the calculation of the FRIACO charge. It assumed that the traffic profile of FRIACO traffic would be the same as PSTN traffic across different LECO links. Oftel believes that this assumption was reasonable in light of the fact that, at the time, there was uncertainty about the likely traffic profiles for FRIACO. One of the key issues in this review is how this assumption is dealt with in light of new information from actual profiles of FRIACO traffic.

3  Measurement issues – timing of Busy Hours

3.1  Before moving on to discuss how the formula for the adjustment ratio, used in the Determination, compares to the correct adjustment ratios under different assumed FRIACO traffic profiles, it is worthwhile considering some measurement issues about the BHs and the EPC.

3.2  The network BH is the hour of the day of greatest demand for the network as a whole. But, when viewed at a more disaggregated level, there is a ‘multiple BH effect’, ie different network components can experience their individual BHs at different times of day. This was discussed in Annex 2 of the Draft Direction, published in April 2000. The multiple busy hour effect is a feature of LECO links, because the composition of customers connected to different concentrators can vary materially. Table 2 below provides a purely illustrative numerical example of the multiple BH effect (note that the numbers themselves are not intended to be realistic).

3.3  Assume, for simplicity, that the capacity on each individual LECO link is dimensioned at an EPC of 0.5. In the network BH (assumed to be time period 3 in Table 2), 390 Erlangs of traffic are observed in total. It might be thought that, given the dimensioning rule, this would suggest that 780 circuits in total would be needed for the 4 LECO links. But, links A, B and D would experience their individual BHs at a different time of day. If the capacity in these links was dimensioned on the Erlangs in the network BH, they would be unable to carry the traffic in their individual BHs. To enable all of the traffic to be carried, 950 circuits would be required. The observed EPC in period 3, the network BH would, therefore, be 0.41 [= 390/950]. But, the EPC measured for each LECO link in its respective individual BHs would be equal to 0.5. The difference is due to the multiple BH effect, i.e. that in the network BH some LECO links are outside of their individual BHs.

Table 2: Multiple BHs on LECO links

4  Theoretically correct adjustment ratio: Simple case – one LECO link

4.1  This section discusses the theoretically correct adjustment ratio. For simplicity, it considers a single LECO link originating FRIACO traffic, which adds to the rest of the traffic on the link, which is referred to as ‘PSTN traffic’. The true constraint is for the LECO link in its individual BH, because at this time of day additional traffic requires additional circuits (capacity) in the link. Adding FRIACO traffic outside the individual BH does not require additional circuits (unless it is sufficiently large when added onto the pattern of PSTN traffic in the link to shift the individual BH to a different time of day). The stylised example below illustrates the point.

Table 3: Adding traffic outside the BH of the LECO link

4.2  FRIACO traffic added in time period 3 would require additional LECO circuits at the rate of 1 ¸ 0.5 = 2 per Erlang, where 0.5 is assumed to be the dimensioning EPC (as in Table 1). In other words the rate of increase of LECO circuits is the EPC of the FRIACO port measured in the LECO’s individual BH, divided by the EPC of the LECO link in its individual BH.

4.3  It is known, however, that different LECO links experience their individual BHs at different times of day (as illustrated in Table 2). Therefore, it is necessary to consider the implications of multiple LECO links before drawing any conclusions.

5  Theoretically correct adjustment ratio: Two LECO links

Introduction

5.1  This is the simplest case that allows for the implications of multiple BHs in LECO circuits. However, even so, a wide range of scenarios are possible reflecting different patterns of FRIACO traffic. As discussed below, scenarios can be constructed in which the adjustment ratio used by Oftel turns out to be the correct ratio, or to understate the correct ratio, or to overstate it.

5.2  The measured EPC in the network BH on LECO circuits can be decomposed into two factors (see Annex B for a further discussion):

• the EPC in each link’s individual BH; and
• the multiple BH effect.

5.3  The EPC on a LECO link in its individual BH it will depend upon the Grade of service required, the density of traffic (because of economies of scale in the Erlang formula), and other factors (such as allowance for resilience and churn). As illustrated in Table 1, the multiple BH effect creates a difference between EPC in individual BHs and the EPC as measured across a number of different links in the network BH.

5.4  A number of illustrative numerical examples are set out and discussed below. The particular figures used in the examples are not based on actual or predicted figures. Rather the purpose is to illustrate a range of possibilities and assist to identify the characteristics that influence the correct adjustment ratio. In each example the starting point is the PSTN traffic taken from Table 2 for LECO links B and C in time periods 2 and 3. Then, FRIACO traffic is added and the effects of different assumptions about FRIACO traffic profiles on LECO circuit requirements are examined. For simplicity, it is assumed that FRIACO Traffic does not shift the busy hour of any link, or the network as a whole. However, since it is not unlikely that the shifting of peaks will occur, its implications will need to be considered in further work.

Two LECO links – numerical examples

5.5  The assumptions made about PSTN traffic, ie before FRIACO traffic is added, are the same in all the following numerical examples and are shown in Table 4. Link B has its individual BH in period 2; and link C its BH in period 3, which is taken to be the network BH. As in Table 2, for simplicity, the EPC in each link’s individual BH is assumed to be 0.5 throughout (i.e. it is assumed not to change even when the level of traffic increases with the introduction of FRIACO). This simplification is made to allow the numerical examples to focus on illustrations of changes in the multiple BH effect. EPCs are shown individually for each LECO link and for LECO circuits overall in each of the two time periods.

Table 4: PSTN traffic assumptions in numerical examples

5.6  The overall EPC in the network BH is the statistic which was used as the denominator in the adjustment ratio in the Determination. Using the illustrative figures in Table 4, it is the total Erlangs on LECO links in period 3 (220) divided by the total number of circuits required (500). The total circuits required for link B depends on the traffic in period 2 and the total circuits required on link C depend on traffic in period 3. Equivalently, the overall EPC is the weighted average of the individual link EPCs, using the number of circuits on each link to give the relative weights.

5.7  In each numerical example it is assumed that there are 30 FRIACO ports (of 64kbit/s each) which, given the FRIACO traffic profile assumptions, require 50 LECO circuits to be added (i.e. by assumption, the sum of FRIACO traffic in period 2 on B and period 3 on C is always 25). The profile of FRIACO traffic differs between the examples in terms of both the traffic on each originating LECO link and in each time period. Since all examples are constructed to result in the addition of 50 LECO circuits for 30 FRIACO ports, the correct adjustment ratio in every example is 50/30 = 1.67. The adjustment ratio given by the formula for the adjustment ratio used in the Determination (and variants thereof) is compared in each example against this correct figure. The illustrative numerical examples are set out at Annex A.

Example 1: Peakier than PSTN traffic

5.8  One line of argument put to Oftel is that the adjustment ratio used in the Determination will understate the correct ratio if FRIACO traffic is peakier on LECO links than PSTN traffic. Example 1 illustrates this concern by assuming that a larger proportion of FRIACO traffic originates in period 2 on link B, which is the time when that link experiences its individual BH (but this is not the network BH). These assumptions about FRIACO traffic involve a lower EPC on the FRIACO ports in the network BH. Taking this as the numerator and the overall EPC for LECO links as the denominator, the adjustment ratio would indeed understate the correct ratio of 1.67.

Example 2: Less peaky than PSTN traffic

5.9  Another suggestion put to Oftel is that the adjustment ratio in the Determination is too high, if FRIACO traffic is less peaky on LECO links than PSTN traffic. The argument is that, with this pattern, traffic is added mostly outside of the individual BHs of LECO links and so can be carried without much increase in capacity. This is illustrated in example 2, in which FRIACO traffic has a flat profile, i.e. the same in both periods for each link. In this example the formula for the adjustment ratio in the Determination does indeed overstate the correct ratio. Essentially, even though the LECO circuit requirement is the same as in all other examples, the FRIACO ports can run 'hotter', i.e. operate at a higher EPC, because of the flatter profile of the traffic by time of day (compared to the profile of PSTN traffic).

Example 3: Same pattern as PSTN traffic

5.10 One rationale used by Oftel to justify the adjustment ratio that it adopted in the Determination was that it would be appropriate if the FRIACO traffic profile were the same as the profile of PSTN traffic. This was not an unreasonable assumption at the time, given the uncertainty about the pattern of FRIACO traffic. Example 3 supports this analysis – the adjustment ratio derived using the formula in the Determination is the correct adjustment ratio.

5.11  The overall EPC on LECO circuits in the network BH is the same before and after adding FRIACO traffic (as is the EPC on each link). The EPC is unchanged because there is no change in either aspect of the multiple BH effect (see Annex B):

• the EPC of link B does not change, because the proportional relationship between traffic on B in its individual BH and the network BH is the same as for PSTN traffic on B; and
• the circuit weights do not change, because the proportional relationship between the number of circuits to be added to B and C to handle FRIACO traffic is the same as for PSTN traffic.

5.12  This result does not require every aspect of the FRIACO traffic profile to be identical to the PSTN traffic profile. Neither of the two characteristics above would alter if a different figure for the Erlangs of FRIACO traffic in period 2 on link C were used. Link C’s individual BH coincides with the network BH, so the amount of traffic outside the BH does not affect the number of circuits required on that link (as long as it is not so large as to shift the peak of C to period 2). Hence, it does not affect the EPC in the network BH on link C, nor the circuit weights.

Example 4: Offsetting effects

5.13  Example 4 shows another case in which the formula for the adjustment ratio in the Determination would be correct. In this example, the FRIACO traffic profile on link B is flatter than PSTN traffic, but peakier than PSTN traffic on link C. The overall EPC on LECO circuits in the network BH is the same before and after adding FRIACO traffic, but this is for a different reason than in example 3. In example 4, the EPC of link B in the network BH increases after the addition of FRIACO traffic. But the overall EPC of links B and C taken together is unchanged, because there is an exactly offsetting effect, namely the reduction in the weight on link C in the average. The change in the weights occurs because the FRIACO traffic assumptions involve more traffic in link B's individual BH (20 Erlangs in period 2) than in link C's individual BH (5 Erlangs in period 3). Variants of this example with slight differences in the assumptions would lead the formula for the adjustment ratio in the Determination to understate or overstate the correct ratio.

6  Implications of the numerical examples

6.1  The factors that affect the correct ratio are:

• the amount of FRIACO traffic per FRIACO port originating on LECO links in the individual BHs of those links; and
• the EPCs of the LECO links in their respective individual BHs.

  6.2  The additional factors that affect the value of the formula for the adjustment ratio used in the Determination are:

• the FRIACO traffic in the network BH (affecting the EPC in the network BH) for LECO links whose individual BH is outside the network BH; and
• the pattern of FRIACO traffic on LECO links' individual BHs compared to the relativities for PSTN traffic (affecting the weights in the overall average EPC for LECO circuits).

6.3  It might be possible to group together into separate categories LECO links that experience the same individual BH. Say that this yielded 4 categories - using the LECO links, A-D, and busy hours, 1-4, in Table 2, the information requirements can be illustrated in Table 5.

Table 5: Illustration of possible information requirements

6.4  In practice, collection of this information and the full-blown calculation of the correct adjustment ratio could be difficult. The question, therefore, is whether an improved formula for the adjustment ratio can be derived which is also practical to implement.

6.5  Comments are invited, in particular, in answer to the following questions:

• What information on FRIACO traffic volumes and patterns should Oftel collect in its Review of the FRIACO charge?
• What practical methodology should Oftel use in the Review to calculate an appropriate adjustment ratio?

7 Comments

7.1  Comments are requested by 11 December 2000. Please send comments to:

Geoff Brighton
Oftel
50 Ludgate Hill
London EC4M 7JJ

e-mail

tel: 020 7634 8925

Annex A: Illustrative numerical examples

Example 1: FRIACO traffic is peakier than PSTN traffic on both LECO links

Example 2: FRIACO traffic has flat profile (less peaky than PSTN traffic) on both LECO links

Example 3: FRIACO traffic has same profile as PSTN traffic

Example 4: Offsetting effects

Summary of effects in the numerical examples

Annex B: Influences on the LECO EPC in the network BH

  The measured Erlangs per circuit (EPC) in the network busy hour (BH) on Local Exchange Call Origination (LECO) circuits can be decomposed into two factors:

• the EPC in each link’s individual BH; and
• the multiple BH effect.

The EPC in the individual BH is assumed to be constant (at 0.5) in the illustrative numerical examples. But, in practice, it will depend upon the Grade of service required, the density of traffic (because of economies of scale in the Erlang formula), and other factors (such as allowance for resilience and churn). The multiple BH effect creates a difference between EPC in individual BHs and the EPC as measured across a number of different links in the network BH.

In the numerical examples link C has its individual BH at the same time of day as the network BH. The EPC of link C in the network BH is therefore also the EPC in its individual BH. But, link B has its individual BH at a different time, so that its EPC is lower when observed in the network BH. The overall EPC in the network BH is the weighted average of the individual EPCs, using number of circuits on each link as the weights. So, FRIACO traffic could change the EPC on LECO links in the network BH for two reasons:-

1. increase the EPC in individual BHs by, for example, increasing the size of routes and so lead to economy of scale gains, ie increase the EPC for a given grade of service due to non-linearities in the Erlang formula (holding constant the time of day profile of traffic);
2. change the EPC measured in the network BH in either direction by changing the multiple BH effect.

Effect i) should be reflected in the adjustment ratio, because it affects the number of LECO circuits that need to be provided per FRIACO port.

There is a change in the multiple BH effect if:-

• the relative weights of links B and C in the overall (weighted average) EPC in the network BH change; or
• the EPC of link B in the network BH changes, which depends upon both the FRIACO traffic on link B in link B’s BH (because this affects the denominator, the number of circuits required) and the FRIACO traffic on link B in the network BH (because this affects the numerator, the Erlangs in the network BH).

These two changes might be in the same or different directions.

In Example 3, there is no change in either factor, leaving the multiple BH effect unchanged.

In Example 4, these two factors move in different directions and are exactly offsetting.

Where, as in Example 2, different LECO links experience their individual BHs at different times of day, if FRIACO traffic flattens the time of day profile of traffic (across the hours in which links experience their individual BHs), it increases the EPC in the network BH (unless offset by changes in the weights). This is because it increases the Erlangs on LECO links which are outside of their BH, without increasing the number of circuits required on these links.

FRIACO traffic reduces the EPC in the network BH if it is peakier across the relevant times of day (unless offset