1 Overview

The main value of the VHN is its connection to the Internet. Even thought the VHN is able to function as an autonomous isolated network the main reason users will deploy a VHN is to access the Internet. The ISP is the link between VHN and the Internet; as such ISP policies have a profound effect on which services can be implemented on the VHN.

The Internet has experienced explosive growth because it is a general-purpose bit delivery mechanism that allows end-to-end communication. The Internet makes few assumptions about the data it carries. New features can be deployed at the edge of the network, without requiring coordination or permission of the network owner. Unfortunately ISPs are trending in the opposite direction, especially the Cablecos. Service offerings are tailored to current usage patterns and restrictions put in place making it hard for new service to be developed. It this situation existed in the early ’90s ISPs would have optimized their network for Gopher and discouraged used of HTML.

This paper discusses business and technical attributes of an ideal ISP and enumerates numerous areas where ISP practices diminish the value of the VHN and are at odds with Internet openness.

The High Tech Broadband Coalition, of which CEA is a member, identified some if these problems. The coalition is lobbying the FCC and Congress to encourage customer friendly regulation.

2 The Ideal ISP

Before discussing restrictive business practice lets examine the ideal ISP service offering from the perspective of the VHN.

2.1 Unlimited IP addresses

A typical VHN will consist of a large number of devices, potentially hundreds of IP nodes. While this may seem ludicrous today one only has to look at the history of electric light and electric motors so see similar adoption pattern. What was once an expensive novelty became ubiquitous.

The ISP reserves a large contiguous address block for each customer. The block needs to be contiguous so each VHN acts as a single subnet. This allows local traffic to remain local. If the address range is not contiguous traffic has to flow across the first-mile connection to the ISP.

2.2 Persistent Addressing

Servers require a stable means of access either by address or name. TCP/IP tends to blur the notion of server. Many IP devices will implement server functions. Most users are familiar with Uniform Resource Locators. DNS translates URLs to IP address. It assumes a persistent IP address that changes infrequently. DNS changes take time to propagate so if the IP address changes the resource is temporally unavailable until DNS is updated.

Persistent addresses are also useful in analyzing traffic and creating access policy rules. Event logs are reported by IP addresses. If the address is not constant interpreting the logs becomes difficult.

2.3 Always on Connection

One of the benefits of broadband is it is “always-on.” This is critical to maximize VHN benefit. It is not sufficient to merely reduce the connect time so it is invisible when an in-home device accesses the Internet. If the connection is not on 24/7 the VHN is not accessible remotely preventing the use of application such as Voice over IP telephony, instant messaging and personal web servers.

2.4 Automatic Configuration

Home user technical expertise varies from technically savvy to novice. Configuring arcane IP address settings is a daunting task for many customers.

Low-level settings such as IP address, subnet mask, gateway address and DNS server address should be performed automatically. Besides easing the end user task automatic assignment allows the ISP to modify these settings automatically as the need arises without having the customer manually effect the change.

2.5 Transparent Transport

The ISP is a transparent bit delivery mechanism. All IETF complaint traffic presented to the ISP is accepted. Unless specifically requested by the customer traffic is delivered on a best effort basis with equal preference for all.

2.6 Speed Tiers

First mile speed constraints will be around for the foreseeable future. A reasonable pricing model is to vary price based on offered speed. This allows the customer to make a price performance tradeoff decision. Once this decision is made service cost is unaffected by usage.

Most residential Internet access is asymmetric, download speed exceeds upload. This imbalance is the result of both technical and business considerations. While there is nothing inherently wrong with asymmetry the difference in download vs upload speed should not be extremely wide or used to effect defacto usage restrictions. Customers should be encouraged to use bandwidth, driving demand, similar to what has be done in the PC industry with CPU speed.

2.7 Quality of Service

Almost all residential broadband suppliers deliver best effort service. This causes problems with data that requires bounded latency, such as Voice over IP. Internal to the VHN various mechanisms can but used with either Firewire or Ethernet to guarantee maximum latency requirement are meet. For the foreseeable future the first mile connection will continue to be the bottleneck between VHN and the Internet backbone. Therefore some form of preferential treatment of latency sensitive data, such as Multi Protocol Labeled Service (MPLS) is needed as long as the first mile connection is unable to delivery adequate bandwidth.

2.8 Authenticated Services

Some ISP services require authentication, such as mail and news. These services should require explicit authentication. This allows the user to authenticate to the service regardless of how they connect. If the ISP authenticates based on IP address if the customer uses a different ISP these services are inaccessible.

2.9 Customer DNS

One of the benefits of always on broadband is remote access to the VHN. This may include public resources like FTP or Voice over IP telephony or restricted to authorized users. Access to the home requires a user friendly URL so users do not have to enter the IP addresses directly. While many residential owners will likely register a personal domain name to decouple naming from that of the ISP the ISP should provide DNS mapping of customer name to IP address. Similar to how ISP based e-mail is handled today. This allows remote access into the VHN by URL rather then IP address. Since the ISP is responsible for changing customer IP addresses they can update DNS with the minimal amount of delay after a change.

2.10 Customer Friendly Acceptable Use Policy

The ISP should not be in a position to determine how and what type of data will be carried over its network. The ISP agrees not to prohibit particular services by blocking TCP/UDP ports or imposing other restrictive policies. This prevents the ISP from over optimizing the network for today’s traffic usage and erecting barriers to new applications.

The ISP is neutral as far as VHN is concerned. Their responsibility ends at the customer premise equipment needed to interconnect to their network.

The scope of the customer’s domain may become a controversial issue, especially with the popularity of wireless networks. Consumer ISP pricing assume typical usage patterns and high neighborhood take rate. Wireless networks threaten to upset that assumption by creating neighborhood ad hoc networks shared over wide area potentially reducing the take rate and increasing traffic.

3 Access Issues

Most ISPs allocate customer addresses from a pool of publicly routable addresses. This means the customer is able to both communicate with remote Internet host and act as a server if they chouse to do so.

Unfortunately some service providers use techniques that limit customers to outbound connections by issuing private IP address or the use of a proxy server.

3.1 NATed ISPs

The IPv4 address shortage encourages ISPs to minimize the use of public IP addresses. One solution is to issue customers IP addresses from RFC 1918 private address pool and use Network Address Translation to convert the address where the ISP connects to the interexchange carrier.

NAT is transparent to most outbound connections but effectively prevent remote access to the customer. The issue is no different then when NAT is implemented at edge of the VHN except the customer does not have to option of specifying mapping rules for inbound traffic, effectively precluding the use of servers.

3.2 Proxied ISPs

Instead of placing customers on private addresses the ISP can force customers to connect through a proxy server, as in the case of AOL. The proxy is able to precisely control how the connection is used preventing use of unauthorized services.

4 How Many IP Address

Due to the address shortage caused by the 32-bit limit of IPv4 addresses, address management has been raised to a fine art. What should have been a simple network identifier has become a valuable and scarce commodity – to be husbanded with great care. The ISP views this scarcity as an additional revenue source.

4.1 Single Address

This is the most common situation for residential customers both dialup and broadband. Each account is issued a single address. Various means have been developed to allow a single addressed to be shared with multiple devices on a LAN.

A common way to share a single address is to allocate private address from RFC 1918 and use Network Address and Port Translation (NAPT) commonly called NAT to translate between private and public address space. Proxy servers are another way to share a single connection and are often used with software implementations. The down side of using a proxy is each application must be aware of the proxy. This limits its usefulness and causes configuration headaches for the user.

Being limited to a single address does not preclude use of “server” functions. Most broadband routers and proxies have the ability to map private services to public addresses.

For example if one wanted to run a web server, the gateway device is programmed to forward all incoming TCP packets for port 80 to the web server. To the remote user the server appears to be at the public address. A limitation of this technique is only a single instance of each device can be used. There is no way to run a second web server since TCP port 80 is already being forwarded to the first server. It is possible to use a different port to establish the connection, but this causes problems unless the remote user knows of the nonstandard port assignment.

As powerful and widespread as NAT is it is not without drawbacks. Protocols like FTP require special processing within NAT at the application level due to the way session ports are allocated. It also breaks some features of IPsec by changing address and port numbers. For more information on NAT refer to RFC 2993 Architectural Implication of NAT

4.2 A Few Addresses

If one desires to operate multiple public servers most ISP charge for additional IP addresses so they are use sparingly. Having multiple public addresses allows multiple servers to be assigned unique IP addresses facilitating access from the outside world.

This is not without side effects. If the addresses are not in the same subnet inter device communication, which would normally be local, is transported over the ISP connection causing congestion and impacting speed. In this scenario NAT is still used for client services to minimize the cost of public IP addresses. This has the same ramifications as a single address, with the added complication of connecting to “public” devices locally.

4.3 Unlimited Addresses

This is the optimum situation but will require deployment of IPv6. Each customer is given a large contiguous address block. This allows each device to have a public address. As long as they are contiguous and part of the same subnet local communication stays local eliminating the degradation caused by forcing the WAN connection to carry traffic that should be local.

In this situation access policy is implemented by a firewall at the edge of the VHN rather then as a byproduct of NAT. This allows the user to control both inbound and outbound access for each device. NAT is often touted as a firewall because without forwarding rules remote access to the LAN is impossible. However NAT has no effect on outbound traffic, so a firewall is still desirable even in a NAT based network.

5 Bridged vs. Routed Network

The most common method used to allocate a “few” addresses to a customer is with a bridged connection. In effect the connection between the user and ISP looks like a LAN. VHN devices are connected to a hub or switch and the broadband modem connected directly to the hub or switch.

Bridged connections operate at ISO layer 2. Devices may be assigned addresses either statically or dynamically. If assigned dynamically the ISP operates the DHCP server. Because bridging works at layer 2 the ISP is privy to the MAC address of each bridged device raising potential privacy concerns. Another downside of bridged networks is access to the DHCP server is lost if the external network fails preventing local use in the event of an outage.

A better solution is to use a routed network. This is the norm for commercial customers but rare for home or small businesses. With a routed network the ISP allocates a block of IP addresses to the customer. The customer manages them as needed. The ISP forwards all incoming traffic bearing the customer’s network prefix to the customer’s router. The router in turn is responsible for deliver within the VHN. This minimizes the effect of remote network failures; the local network still works correctly of the ISP fails and it hides VHN specific information from the ISP.

6 Address Allocation

IP addresses may be assigned statically or dynamically. Static allocation is communicated between the ISP and customer though an out of band channel. The customer is responsible for configuring VHN devices. If the ISP needs to change this information the customer has to be notified and the changes updated manually. Dynamic assignment is much easier for the ISP to manage. Automatic mechanisms are used to issue addresses on an as needed basis, and if necessary the ISP is able to rebalance address utilization without involving the customer.

From the customer perspective a static address is more convenient for running servers. A static address provides a persistent address facilitating access by remote hosts.

For most home user the best of both worlds is a pseudo-static address as implemented by many Cable ISPs. The allocation mechanism is dynamic but bound to the MAC address of the user’s device. This means the customers address stays fixed for long periods of time, while still giving the ISP the ability to change them automatically on an as needed basis.

7 Encapsulation (PPPoE/PPPoA)

The Point-to-Point-Protocol is used to facilitate sharing the first-mile network by multiple ISPs. In effect the network is transformed into a virtual point-to-point connection between the customer and ISP. The ISP authenticates the customer, using the same RADIUS mechanism used for dialup. While purists cringe at the use of PPPoE Point-to-Point-over-Ethernet and PPPoA Point-to-Point-over-Asynchronous Transfer Mode (ATM) it is an effective tool to share a common physical network.

The down side of PPP is that it is an encapsulation protocol, each packet requires 8-bytes of overhead. This in and of itself is not a significant performance issue because maximum Ethernet packet size is 1500 bytes. What does cause problem, beside poor software implementation, is the reduction in maximum packet size. Applications are supposed to verify end-to-end limits on packet size. If this is not performed correctly and a 1500 byte packet is created when it is passed to the PPP layer the addition of 8 bytes will either cause the packet to be fragmented, impacting speed or be rejected preventing communication.

PPP is commonly used in conjunction with dynamic IP assignment, but it can also be used with static addresses.

A little known feature of PPP is that it allows the negotiation of multiple PPP sessions. Some carriers see this as a way to sell value add services bypassing the ISP. Separate PPP connections are set up for say video streaming or digital telephony creating a virtual PPP connection between the customer and service provider. This has profound effects on VHN because it creates additional access methods that may or may not be IP based. It is also very much at odds with the notion of Internet end-to-end connectivity with services implemented at the network edge.

8 Asymmetric Speed

First mile access speed will likely be a precious commodity for the foreseeable future. One of the ways providers address customer preferences is tiered service. Faster speed costs more Carriers have limited ability for downward price flexibility since must of the cost of broadband access is independent of speed. The motivation of tiered price is crude control of bandwidth consumption. This is especially important for Cable networks since each user shares a fixed medium with neighbors. The only was to provide acceptable service is to reduce the number of subscribers on a segment or reduce modem speed.

This is a reasonable mechanism for providers to segregate customers by willingness to pay.

Providers are using asymmetric speed to control how customers use the network. Due to technical and business consideration both Cable and DSL speed is asymmetric, download is considerably faster then upload. This is not unreasonable given the bulk of most residential traffic is weighted toward downloading. However an unfortunate side effect of this optimization is the growing emphasis on the residential market as a data sink rather then a network peer both “sourcing” and “consuming” information. This optimization bodes poorly for the emergence of cooperative distributed networks where data is distributed rather then located in centralized servers.

Asymmetric connection speed should not be used to thwart innovative use of the Internet and relegate the VHN as an information sink.

9 Latency

The notion of latency and bandwidth are often confused. High-speed connections can also experience high latency. A truck full of CD-ROMS is a good example. It takes a long time for the truck to arrive – high latency - but once it does it delivers tremendous bandwidth. More realistic is latency introduced by satellites in geosynchronous orbit. Another problem area is data interleaving error correction used by DSL this minimizes errors but ads substantial latency.

The most critical service from a latency standpoint is real time voice telephony. Round trip latency over 200 ms degrades quality and latency above 500 ms makes conversation virtually impossible.