Impact of the Net Neutrality Repeal on Communication Networks: Profit-Driven Model Results

Table of Links

Abstract & Introduction

Related Works

Repealing Net Neutrality

Profit-Driven Model Results

Conclusions & References

Biographies

IV. PROFIT-DRIVEN MODEL RESULTS:

In this section, we evaluate the increase in ISP profit and the reduction in network traffic and subsequently power consumption resulting from the optimized pricing scheme under the repeal of net neutrality. We define the three services classes as follows;

• Class A; for UHD video service; 18 Mbps download rate.

• Class B; for HD video service; 7.2 Mbps download rate.

• Class C; for SD video service; 2 Mbps download rate.

We investigate CP’s end users’ choices of service classes based on different PED. We show how users behavior under the different PED; 0.2, 0.4, 0.6, 0.8, 1 or 2 affects the equilibrium price of each class the ISP charges the CP for delivering its content.

As discussed above, we assume that the CP will transfer the price increase to their customers at the same rate (if the CP absorbs some of the increase in prices, then this may represent a different PED). As a benchmark, we consider users to be distributed among classes according to the Cisco forecast report [45], where UHD, HD, and SD users distribution are 19%, 56% and 25% respectively. We consider 1.8 million users active simultaneously in the network. This figure is obtained as follows: The number of users is 44 million users in Netflix in the US and the average user spent around 1 hour per day watching movies in 2015 [46]. Therefore, the average number of users during one hour of the day is 1.8 million users, which is an average number that does not consider the popularity of different viewing times in the day. The concentration of users at any node in AT&T network is based on the population of the state where the node is located (see Fig. 2).

We consider the BT network connectivity selling price as the net neutrality price of the three classes where 10 Gbps connectivity is priced at £12,600 ($15,750) per year [47], i.e. $131 per 1 Gbps link per month. The actual cost of provisioning ISP core network infrastructure is sensitive information and not usually shared by ISPs. However, we estimate the cost of provisioning 1 Gbps of network as $118 considering 10% as the ISP profit margin (the average profit margin for AT&T [2] and Comcast [48] were approximately 9% and 12%, respectively between 2013-2018). We divided the cost among the three network layers; core, metro and access network based on their power consumption percentages: 24%, 6% and 70%, respectively [49] which corresponds to $28, $7, $83, respectively. The cost of $28 per Gbps in the core network is associated with a single hop. For the AT&T architecture the average hop count between clouds and other nodes is 1.

As shown in Fig. 2, we choose AT&T core network (a primary core network topology in the US) as a core network topology example. This core network consists of 25 nodes and 54 bidirectional links. AT&T hosts datacenters in nodes 1, 3, 5, 6, 8, 11, 13, 17, 19, 20, 22, and 25 [50]. These nodes are used to host datacenters to serve distributed CPs users. The input parameters used are given in Table I.

In the following subsections, we evaluate two scenarios; equal PED for all classes and different PED for different classes.

Under each scenario we study three scenarios of delivering CPs contents to users; a cloud-based delivery and a cloud-fog based deliver and fog-based delivery.

1) Equal PED among classes:

In the following, we study three scenarios of delivering CPs contents to users; a cloud-based delivery and a cloud-fog based delivery and fog-based delivery.

Cloud based delivery: Figs. 3 to 5 show the profit-driven model results for AT&T core network where content is delivered from the 12 datacenters in the AT&T topology [50]. The number of users and the corresponding price of each class under different PED are illustrated in Fig. 3. The primary y-axis shows price per Gbps per month of each class in US dollar. These prices represent the equilibrium point of users’ willingness to follow the price increase which results in maximum profit for the ISP. The secondary y-axis corresponds to the percentage of users subscribed to each class. The x-axis shows different PED scenarios from 2 to 0.2. The former represents the highest sensitivity to the price change considered, whereas, the latter represents the contrary. PED values are shown along with the case of net neutrality where the price of different classes is fixed at 113$ and the percentage of users in each class follows Cisco forecast report [45] as discussed above. For each PED value we consider two cases; a case where the optimized pricing scheme should maintain 100% of the users that existed under net neutrality (LB ≥ 100) and another case where the pricing scheme can result in users leaving the service (LB ≥ 0).

Fig. 4 is a plot of the monthly profit of ISP considering different PED values as well as net neutrality scenario. Total traffic of core network and the power consumption due to this traffic under different PED scenarios and the net neutrality scenario are plotted in Fig. 5. In case of content with PED = 2, under LB ≥ 100 or LB ≥ 0 , Fig. 3 shows that repealing net neutrality has increased class C users to 48% of the total number of users compared to 14% only under the net neutrality pricing scheme. This increase is a result of some users of class B downgrading to class C as the class B price increased slightly by 18% (the number of users in class B reduced to 36%) and due to new users joining the service (the total number of users increased to 102%) attracted by the 1% decrease in class C price. The users of class A are reduced to 18% of the total number of users as a result of the slight increase in price by 19%. This pricing scheme and distribution of users have resulted in an increase in the total profit by 54% compared to the net neutrality scenario as seen in Fig. 4. For a less sensitive content with PED = 0.2 under LB ≥ 0, the equilibrium pricing scheme resulted in 28% of the users leaving the service as the increase in the classes price resulted in an increase in the profit by a factor of 8.3 compared to the net neutrality scenario. Maintaining all the users of the service (LB ≥ 100) has slightly reduced the profit by 10%.

In addition to growing ISP profit, we also observe in Fig. 5 a decline in the core network traffic by up to 55% under PED = 0.2, LB ≥ 0 and a consequent reduction in power consumption by 49%. This reduction in core network traffic and power consumption occurred for two reasons; 1) some cloud service users leave classes A and B to subscribe to class C as the charges per Gb/s of the classes A and B increase. 2) the total cloud service subscribers diminished due to the increase in class C price (in case of LB ≥ 0).

Cloud-Fog based delivery: Next, we introduce 10 fog nodes in addition to the 12 datacenter locations. These fog nodes are assumed to be built in the proximity of nodes with the highest population in the AT&T core network, so no core network cost (Ͼ) is incurred by serving the demands of these nodes. Fig. 6 shows that the prices per Gbps per month under different PED that are less than the previous case (cloud-based delivery) as we reduced the cost of the core network by introducing the fog nodes. Under PED = 2, the prices compared to the net neutrality case in class A and B increased by 12% and 11%, respectively, while the price of class C dropped by 1% as opposed to 19%, 18% and 1% with cloud-based delivery. The reduced prices attracted more users resulting in increase in the profit by 18% compared to the net neutrality case as seen in Fig. 7 as opposed to a 54% increase in profit with cloud-based delivery. Fig. 8 shows a reduction in core network traffic (40%) and power consumption (35%) by repealing net neutrality in the cloud-fog architecture.

Fog based delivery: Here, we consider a scenario in which all users access CP contents from a local fog node. Although deploying a fog node locally, to serve CP customers, increases the capital expenditure (CAPEX) and operating expenses (OPEX) of provisioning multiple locations (i.e. 25 fog nodes in AT&T network), it reduces the communication network transit cost burden to the minimum. However, fog nodes are not always an option due to the finite capacity of processing and storage. The results show that the prices are further reduced under fog-based delivery (Fig. 9) as no core network cost (Ͼ) is incurred by serving demands. For instance, under PED = 2, the prices compared to the net neutrality case in class A and B increased by 9% while the price of class C is decreased by 11% resulting in increase in the profit by 6% compared to the net neutrality scenario as seen in Fig. 10.

2) Different PED among classes:

In this section, we consider a scenario where elasticity of demand varies among the different classes of service. We consider class C to be less sensitive to price change than class B. Also, we considered class B to be less sensitive than class A. The elasticity of demand for classes A, B and C are considered to be 2, 0.8 and 0.2, respectively. Fig. 11 shows the price per Gbps for classes A and B is the same under different scenarios and delivery schemes as a result of the high PED of class A. Class C is priced at the same level of classes A and B for LB ≥ 0 as the low PED of class C limits the number of users leaving the services as a result of increase in the price. Fig. 12 shows an increase in profit by up to 88%, 29% and 16% under cloudbased delivery, cloud-fog based delivery and fog-based delivery, respectively, compared to the net neutrality scenario. Fig 13 shows a decrease in core network traffic by up to 43% and 30% under cloud-based delivery and cloud-fog based delivery, respectively, compared to the net neutrality scenario. Also, the total reduction in the core network power consumption (as shown in Fig 14) is up to 40% and 32% under cloud-based delivery and cloud-fog based delivery respectively.