1 Broadband ISP Overview

Internet popularity is driving demand for ever-faster service, and exerting downward pressure on price. Connection between end user and ISP is often called the last-mile. This implies there is a magical entity out there called “The Internet” and customers are passive consumers of Internet goodness. I prefer the term first-mile. It better denotes Internet value being the result each person’s connection as both contributor and consumer. Today most citizens in industrialized countries have access to some form of high-speed access and more is coming on line every day. Broadband is increasing being seen as a utility without which citizens are unable to fully participate in society.

Broadband is a much abused and inexact term. United States Federal Communication Commission (FCC) is constantly redefining minimum broadband speed. It had been 200 kbps. Basic broadband increased requirement to 768 - 1500 kbps toward customer (downstream). As of July 2010 the National Broadband Plan increased minimum speed to 4Mbps toward customer and 1 Mbps up.

Most of us utilize an Internet Service Provider (ISP). ISP owns leases or otherwise has access to a data path to each customer. The picture at left provides a high level overview of how ISP connects customers to the Internet. Diagram is somewhat dated in that 1.544 Mbps T1 links would be something faster today.

Connecting to an ISP would not have much value if the only people you can communicate with are other customers. To provide worldwide connectivity ISPs connect to other ISPs at peering points. This allows traffic be delivered anywhere in the world.

ISPs exert a great deal of control over how customers use the Internet. Much is made of Internet robustness and redundancy. That is true of the Internet in general but for most of us the ISP controls the Internet on ramp and acts as gatekeeper. In most locations broadband competition is nonexistent or extremely limited. ISP business policy has significant impact on how customers use the Internet and how new services are deployed.

There are several essential functions that must be provided by the ISP, as they are the only entity capable of doing so. There are many services, often associated with ISPs, which can be provided by anyone. The distinction between essential and non-essential functions is important when discussing Network Neutrality. As broadband access becomes more pervasive ISPs need to balance business considerations with public interest.

Essential Core Functions

· Customer Connection

· Authentication

· Address Allocation

· IPv6 support

· Packet Routing

· Peering

· Multicast (IGMP)

· Quality of Service (QoS)

· Service Level Agreement (SLA)

· Acceptable Use Policy (AUP)

· Support

· Billing

Non-Essential Services

· Virtual Private Network (VPN)

· Name Resolution (DNS)

· E-mail

· Usenet

· Web Hosting

· Voice over IP (VoIP)

· Fixed Mobile Convergence (FMC)

· IP Radio

· IP Television (IPTV)

1.1 Essential Core Functions

Core functions must be provided by the ISP, no other entity is able to provide them.

1.1.1 Customer Connection

ISP provides either a routed or bridged customer connection. Residential accounts are typically bridged; customer connects to ISP as if they were part of the ISP LAN. VLAN techniques prevent users from seeing each other’s traffic. Business class accounts are typically routed rather than bridged. ISP’s edge router communicates with customer’s edge router. Routed connections are more flexible, but also more complex, then bridged.

Some ISPs own the First-Mile access network; Cable and FTTP service are examples of this type of ISP. Some ISPs rent physical access. This is typical with DSL where copper circuit is rented from Incumbent Local Exchange Carrier (ILEC). Dialup ISPs use Public Switched Telephone Network PSTN to connect customers. Wireless ISPs do not provide a physical connection at all other than obtaining an FCC license to use the public airwaves.

Customer interface requirements differ greatly depending on type of service and whether or not ISP provides network access device. For example Cable, DSL and FTTP ISPs typically provide customer with standard’s based modem with either Ethernet or USB as customer interface. In the US T-1 is a tariffed telecommunication service. The FCC defined customer interface as two pair copper circuit typically implemented by a smart-jack. Dialup ISPs require customer obtain a V.90/92 or ISDN modem. Wireless ISPs typically supply and install customer antenna and radio.

1.1.2 Authentication

ISP needs a mechanism to insure only authorized customers connect to its network. For some types of service link between customer and ISP is hardwired so any traffic appearing on the link is assumed to originate from customer. T1, SDSL and FTTP are typical hardwired connections. Shared media such as Cable and wireless need a way to identify customer. DOCSIS modems include digital signatures to prevent unauthorized access. ADSL ISPs typically use Point-to-Point Protocol over Ethernet (PPPoE) to authenticate customers. Telco’s like PPPoE because it facilitates support for third-party ISPs. Dial up ISPs typically utilize Point-to-Point Protocol (PPP) to authenticate customers using same RADIUS servers as PPPoE.

1.1.3 Address Allocation

Each Internet host requires a unique address. ISPs typically provide residential customers with a single IPv4 address. Large customers may obtain their addresses directly from Internet Corporation for Assigned Names and Numbers (ICANN) or from wholesale ISPs. IPv4 defines a 32-bit address space yielding about 4-billion possible addresses. That was a large number back when the Internet was limited to educational and government institutions but has become a serious limitation today. As a result IPv4 addresses are in very short supply. Next generation Internet protocol IPv6 increases address space to 128-bits, a truly humongous number. With IPv6 even residential customers are issued a large block of addresses.

IP addresses serve multiple functions. They denote a specific Internet host; each host needs at least one IP address. IP addresses also facilitate routing because they are allocated in blocks. If IP addresses were issued randomly each router would need to potentially look through billions of addresses to determine how to handle each packet. By aggregating addresses into large blocks routers only need look at a few high order address bits to determine how to forward packets.

Business accounts are typically configured statically. Static allocation is preferred for commercial accounts. With a static address customer settings are configured manually, based on information provided off-line by the ISP. This eliminates possibility of address change interfering with remote access.

Most residential accounts obtain IP address dynamically. This is convenient because it eliminates need for non-technical customers to manually configure IP address, subnet mask, gateway address and DNS server address. Dynamically assigned address may change at any time making it difficult to operate servers.

1.1.4 IPv6 Support

February 2011 witnessed a major milestone on the journey to mass deployment of IPv6, IANA made the final allocation of IPv4 addresses. This event has been long anticipated but having finally occurred ought to spur more rapid deployment of IPv6, the successor to IPv4. IPv6 represents a significant improvement over IPv4 but adoption has been painfully slow. The reason is that IPv6 is not backward compatible with IPv4. This is because while IPv4 uses 32 bits for address supporting approximately 4 billion hosts (4.3 x 10⁹⁾IPv6 uses 128 bits for a mind boggling 340 Undecillion hosts (3.4 x10³⁸⁾. The massive address space allows large blocks of address to be allocated, thus easing routing and management.

Since IPv6 is not backward compatible ISPs offer a number of ways to support the transition.

1) Dual-Stack is probably the easiest to understand. The ISP provides customer with both IPv4 and IPv6 addresses. Customer’s equipment uses the appropriate version to communicate with the remote host. The down side of this implementation is the need to provide the customer with a routable IPv4 address.

2) Dual-Stack lite The ISP provides only IPv6 address. All traffic between customer and ISP network is IPv6. The ISP uses NAT so multiple customers are able to share a single IPv4 address. Much like the way a typical home network shares a single IP address today. Customer’s router disencapsulates IPv4 packets and distributes IPv4 and IPv6 packets appropriately within the LAN.

3) Tunneling (6in4) is a way for IPv6 packets to be transported over an IPv4only network to another IPv6 network. This is probably not of interest for most of the readers of this paper but is very useful for companies with many locations.

4) Translation enables IPv6 only hosts to communicate with IPv4 hosts. The ISP provides Network Address Translation (NAT) to convert IPv6 address to IPv4. This is made easier because the IPv6 address range is larger than IPv4 it completely subsumes IPv4 address space.

For more background check out: IPv6 at home – a guide to getting started.

1.1.5 Packet Routing

The term Internet is a contraction of Inter network. Internet is literally a network of networks. Routers are used to forward packets between networks. Devices on the local network know if host they are trying to reach is local or not. To access a remote host packet is forwarded to a router attached to the local area network LAN. The router uses its knowledge of connection topology to make intelligent forwarding decisions. This process is repeated multiple times until packet finally reaches its ultimate destination. Routers learn connection topology by exchanging routing information.

1.1.6 Peering

Signing up with an ISP would not be very useful if customer was limited to only communicating with other customers. The early Internet consisted of a few nodes interconnected by point-to-point links rented from the old Bell System. As the Internet grew it became apparent there was a need for a high-speed data network to interconnect high usage nodes. Transit providers span continents and oceans providing the backbone. Peering allows transit providers to exchange traffic with each other and accept end user traffic. Large companies, ISPs and governments connect directly to peering points. Smaller ISPs purchase bandwidth from third party wholesale suppliers. The end result is regardless how one connects it is almost always possible to communicate with anyone else on the Internet.

1.1.7 Multicast (IGMP)

Internet is a powerful communication medium. A user is able to connect to another anywhere in the world virtually instantly. As powerful as this type of communication is it is not well suited for broadcast, delivery of one program to many subscribers simultaneously. Traditional broadcast business model grew out of the technical limitation of radio. Station owner built a transmitter and anyone within range was able to receive the broadcast.

The one-to-one connection model used by the Internet makes it difficult to cost effectively broadcast programs since each listener requires a unique network session. Internet Group Management Protocol (IGMP) creates the infrastructure to deliver a single stream to multiple users. At each branch a decision is made whether or not to forward the stream. If an active listener is downstream packets are forwarded, if not they are dropped. This conserves channel capacity by suppressing streams no one is listening to. IGMP dramatically reduces server load since only a single copy is transmitted. Internet broadcasting is still in its infancy and IGMP is not commonly implemented by ISPs. For multicast to function each router between sender and receiver needs to support IGMP.

1.1.8 Quality of Service (QoS)

Internet is an egalitarian best effort network. This works amazing well for transferring large chunks of data from point A to point B. The network continues to operate in the presence of all sorts of impairments and failures. However: best effort does not work as well with latency critical applications such as telephony and streaming media when dealing with congestion. For example a Voice over IP (VoIP) phone call requires round trip latency under 150ms. Excessive delay makes carrying on a conversation difficult and when extreme virtually impossible. On the other hand if a print job is delayed a little no one is likely to notice as long as it completes successfully.

When a switch or router encounters congestion it buffers incoming packets until it is able to forward them. Normally this occurs on a first in first out (FIFO) basis. Quality of Service (QoS) metric allows latency sensitive packets to receive priority queuing. This simple strategy works well if latency critical traffic is a small percent of total. QoS marks packets with a (Diffserv) priority level. When congestion occurs higher value packets are delivered first. Lower value packets are delayed or discarded during periods of extreme congestion. QoS service allows more graceful degradation by moving high priority packets to the head of the queue.

As discussed in later section traffic shaping and preferential packet treatment is controversial. Network Neutrality proponents are concerned ISPs will strike business deals with partners to preferential deliver their data at the expense of competitors. It is important to remember Quality of Service mechanisms do not provide additional channel capacity. They simply redefine winners and losers. When channel capacity does not meet “offered load” (an old telecom term) some policy must be in place to deal with congestion. The PSTN managed congestion by withholding dial tone or returning an “all trunks busy” message when call could not be completed. The Internet handles congestion by delaying packets or in extreme cases dropping them. QoS controls which packets get delayed. Many argue deploying additional capacity is more cost effective then implementing a complex differential service mechanism.

To be maximally effective QoS requires end-to-end deployment. Technical and business problems facing QoS is much the same as IGMP. There is little value until “everyone” deploys it and little incentive to be an early adopter. ISP and all intermediate nodes need to monitor packet privilege level and treat them accordingly. Controls at each level need to monitor statistics to prevent “tragedy of the commons.” If too many packets ask for priority handling they all suffer.

Most residential broadband service is asymmetric; download is much faster than upload. There is benefit in shaping upload traffic so higher priority traffic is treated preferentially at the edge of the customer’s network. Customer’s edge router examines outbound packets and prioritizes them. Many residential routers already do this to a limited extent giving TCP/IP ACKs get preferential treatment. Similar treatment may be applied to VoIP or critical gaming packets.

1.1.9 Service Level Agreement (SLA)

One of the main differences between residential and business accounts is the Service Level Agreement (SLA). SLA defines things like: minimum speed, maximum latency, service reliability and mean time to repair. SLA guarantees are one of the reasons business class service is so expensive. Residential accounts are best effort. If connection fails or experiences congestion ISP is under no obligation to correct problem on an expedited basis. SLA imposes performance guarantees ISP must meet and penalties if they do not.

1.1.10 Acceptable Use Policy (AUP)

Acceptable use policy (AUP) defines customer responsibility, how service may be used and penalty for misuse. For example, residential customers are typically prohibited from reselling access or running servers and ISP often block certain types of traffic. In an attempt to reduce cost some residential ISPs impose usage caps to limit monthly download and upload. Most ISP’s reserve the right to revise AUP at any time making for a pretty one-sided contract.

1.1.11 Support

Regardless of how good service is on occasion will be necessary to contact technical support to resolve problems. Tech support responsiveness dramatically affects overall customer satisfaction.

Due to hyper competitive nature of residential broadband most providers offer only limited help in troubleshooting problems. Finger pointing can be frustrating when a customer is trying to resolve a complex interaction and ISP does not consider it their responsibility. Specialized web sites such as Broadband Reports can be an effective alternative. Broadband Reports is a good example of an Internet community, members post questions and assist each other in dealing with network issues.

1.1.12 Billing

ISPs would not stay in business long if they could not bill for service. Most ISPs offer flat rate billing based on speed tiers. Some ISPs set monthly bandwidth consumption quotas. Exceeding quota results in extra (sometimes substantial) charges or a reduction in speed. During the Dotcom era many dialup ISPs offered advertising supported free access. Most of these companies are long gone.

There is no comparable notion of telephone long distance in the Internet world. It does not cost any more to send a message across the street as around the world.

1.2 Non-Essential Services

This section examines services often provided by ISPs but that can be provided by third parties and in some cases even the customer. This distinction is important in the Network Neutrality debate. If an ISP decides to offer a non-standard or value-add service and customer or third party is able to supply similar service the impact is dramatically different then if the ISP implements proprietary core services.

1.2.1 Virtual Private Networking (VPN)

Virtual Private Network (VPN) uses public Internet to create private communication paths. Depending on how it is implemented it may be a feature that only the ISP is able to deliver or something customer or third-party is able to engineer.

Large companies make extensive use of MPLS to implement a geographically dispersed corporate LAN. To users, regardless of location, resources appear to be on the LAN. Service provider configures edge routers such that data presented to it is delivered to the correct physical location. ISP isolates each company’s traffic so in is invisible to other companies.

More MPLS/VPN details in this Network World article.

It is also possible for customers to create their own VPN using IPsec. In this case customer, rather than service provider, creates a secure end-to-end path through the public Internet. IPsec is used extensively to support satellite offices and telecommuters.

SSL/TSL is another mechanism used to provide end-to-end privacy. SSL was originally developed by Netscape to protect web based financial transactions. Because it is built into all browsers many companies are using it, rather than IPsec, to provide remote employee access.

1.2.2 Domain Name System (DNS)

The Domain Name System (DNS) translates Uniform Resource Locator (URL) to IP address. Without DNS web sites would have to be accessed by IP address. DNS is unique in that it is the only fully distributed database in existence. DNS name space is evaluated right to left. Naming convention begins with an implied “.” at the extreme right of the top level domain (TLD), the root level domain. Next in the hierarchy are the TLDs (com, gov, edu, uk, ru), then registered domain name (tschmidt is my registered domain within the .com top level domain), than one or more sub domains. As each level is traversed it provides information about then next lower level until the IP address of the particular host is determined.

If DNS is unable to resolve a domain name it returns an error message. Some ISPs have attempted to monetize entry of incorrect URLs by returning advertising supported web page if URL cannot be resolved. DNS redirection is controversial. Some customers may find redirection useful, other not. For those who do not it may be possible to disable this feature or run their own DNS Resolver. For those who want to run their own DNS Resolver TreeWalk is popular

1.2.3 E- Mail

As with DNS just about all ISPs provide email. It is wise to consider ISP e-mail account a throwaway. If you change ISP or the ISP is sold email address changes making it difficult for folks to stay in touch. For a more permanent address use one of the free e-mail services such as Yahoo or Gmail or better yet register your own domain.

1.2.4 Usenet

Usenet Newsgroups are a valuable source of up to date information. Usenet is text based and predates the web. Most ISPs used to include Usenet access. Due to declining interest in Usenet and legal attacks related to pornography many ISPs are taking easy way out and eliminating or scaling back support of Usenet. Usenet access is available from a number of specialized companies. Newsadmin has a nice comparison list of newsgroup providers.

1.2.5 Web Hosting

Many ISPs provide web site hosting for residential customers. This allows customers to have an Internet presence without need to register a domain name or run their own server. ISP runs a virtual server enabling many web sites to run on a single computer. ISP web hosting is a boon to residential customers by providing a painless way to create web presence. As with email use of the ISP web server binds customer’s web site to the ISP. There are many hosting alternatives that decouple personal web sites from the specific ISP.

1.2.6 Voice over IP (VoIP)

Public switched telephone network (PSTN) represents a hundred years of engineering. Recently packet based telephony has become a serious contender. Rather than traditional circuit switching Voice over IP (VoIP) uses packet-based communication to deliver two-way real time voice. Voice communication is very demanding. Voice data rate is low by Internet standards only 8-64 kbps in each direction. However latency is critical. If packets are delayed more than a few hundred milliseconds voice quality is seriously degraded.

As with any new technology many players have entered the market. Most will fail but a few will succeed. If your ISP offers VoIP check the service thoroughly. The asymmetric nature of most residential service, upload being much lower than download, makes it easy to saturate the connection. Some form of Quality of Service (QoS) may be required to mark VoIP packets, as high priority so they get preferential treatment.

1.2.6.1 Number Portability

In the US FCC mandated telephone number portability. In most cases you will be able to transfer your existing wire-line phone number to new VoIP phone service.

1.2.6.2 E911

E911ememgency service can be difficult to implement with Voice over IP. There is no easy way to physically locate the caller. Unlike wire-line POTS where location of the phone never changes, a VoIP call can originate anywhere. Cellular networks have struggled for years to implement E911 service using triangulation or GPS to locate subscribers.

1.2.7 Fixed Mobile Convergence (FMC)

There is tremendous interest in multimode cellular phones able to utilize both traditional cellular network and opportunistically, Wi-Fi networks. Fixed Mobile Convergence (FMC) represents a win-win situation for both customer and wireless provider. For providers it utilizes the vast potential of the Internet and private LANs to remove traffic from expensive cellular radio networks. For customer it represents potentially lower cost and better performance. For business it represents a way to eliminate traditional PBX wired telephone infrastructure without paying extravagant per minute charges. Depending on legal restrictions it may offer arbitrage advantage for multinational corporations to treat voice like email, bypassing local phone companies and eliminating per minute charges.

An alternative to Wi-Fi is femtocells being offered by several Cellular phone companies. Femtocells are low power cellular base stations that utilize customer’s broadband connection to deliver coverage to a single home. As with Wi-Fi Cellular providers like it because it moves traffic off cell stations.

At this stage it is unclear as to ISP’s role in FMC. Theoretically any packet-based network can be used to transport voice. However, Cellular providers may be reluctant to have just anyone connect directly to their network. They may want to limit access to a few chosen providers. Only time will tell.

1.2.8 IP Radio

ISPs do not appear much interested in becoming content aggregators for radio the way they are for TV. But other than much lower bandwidth Internet radio is not all that different then Internet TV. Radio-Locator is a convenient way to find Internet radio stations.

1.2.9 IP Television (IPTV)

Over-the-air (OTA), Cable and DBS TV all use basically the same transmission scheme. Individual programs are assigned a channel within the allocated RF spectrum. US TV channels are 6 MHz wide, in Europe 8 MHZ. Channels were initially specified to carry a single analog standard definition TV program. Migration to digital TV allows each channel to carry a high definition program (HDTV) or multiple standard definition programs (SDTV).

IPTV represents a fundamentally different way to deliver TV leveraging packet-based technology. IPTV opens the door to demand based programming. Instead of changing channels video on demand (VoD) is more like going to the library. One simply selects the program of interest and it is delivered instantly anywhere anytime to any device the end user chooses. Using MPEG-2 compression SDTV requires about 2 Mbps and HDTV 15 Mbps. MPEG-4 yields significantly lower data rates for equal image and sound quality. These rates are the result of spectral (within the picture) and temporal (over time) data compression. Raw data is much too high to be delivered economically.

Historical residential ISP assumed a customer traffic model of primarily bursty download traffic such as loading web sites or email. Streaming TV and to a lesser extent streaming radio lock up significant bandwidth for extended periods of time. This is much more demanding then browsing.

IPTV dramatizes the disruptive nature of the Internet. Since the end of WWII Cable companies have wired areas to deliver broadcast TV over coax and more recently fiber. Cable network is intimately bound to TV delivery. As residential broadband speed increases the door opens for new providers to bundle content and deliver it without the need to either build or own the means of delivery. ISPs are worried about being relegated to commodity bandwidth providers. As such they are promoting triple-play: data, telephone, and TV as a way to increase revenue per customer and lock in customers.

1.2.9.1 Content Delivery Network

Distributing video is very bandwidth intensive. Demand based video requires a one-to-one connection between user and server as opposed to the one-to-many connection needed for broadcast. Being demand based each user may be viewing a different program or different time within the same program. To address the growing demand for video specialized service providers, called Content Delivery Network (CDN), have become popular. The CDN replicate programs on many servers and locates them near the ultimate end user. Often they have special peering arrangements with large ISPs. CDNs reduce the amount of traffic flowing over the Internet backbone because they are able to source the file near where it is being viewed.

1.3 Connection Sharing

Historically when a customer contracted with an ISP they were given a block of IP addresses large enough to meet their needs. IPv4 address shortage forced ISPs to rethink how they allocate scarce addresses. Most residential broadband ISPs restrict customer to a single IPv4 address. This creates a quandary; how to cost effectively connect multiple hosts to the Internet? The most common workaround is Network Address Translation (NAT) coupled with use of private IP addresses. RFC 1918 reserves three blocks of IP addresses guaranteed not used on the Internet. Because these addresses are not used on the public Internet they can be reused multiple times.

Combining NAT, more properly Network Address Port Translation since both address and port number are modified, and private addresses allow a virtually unlimited number of computers to share an Internet connection even though ISP only provides a single address. NAT provides translation between private addresses on LAN and single public address issued by ISP on WAN.

LAN traffic is not affected by NAT. When a request cannot be serviced locally it is passed to NAT router, called a gateway. Router modifies packet by replacing private address with public address issued by ISP and if needed modifies port number to support multiple sessions and calculates new checksum. Router sends modified packet to remote host as-if-it-originated-from-the-router. When reply arrives back at the router modifications are reversed and packet is forwarded to host on the LAN. Router tracks individual sessions so multiple computers are able to share a single address. From the Internet perspective local hosts are invisible. The router looks like a single computer with the address of the public IP issued by the ISP.

IPv6, with its vast address range, does not require NAT. Each device will have its own public IP address. This changes the nature of residential routers. NAT, though not technically a firewall, blocks all incoming connection requests from remote hosts. Unless specifically programmed with port forwarding rules it does not know which device on the LAN to forward the request. This default behavior is lost with IPv6. Residential routers that support IPv6 should block incoming connection requests unless specifically programmed otherwise.

1.4 Blocked Ports

Internet is designed as a transparent end-to-end bit delivery network. This means any host is able to communicate with any other host. TCP/IP and UDP/IP use ports so host can manage multiple simultaneous sessions. Ports are 16-bit unsigned values yielding up to 65,535 ports for each connection type. When a service is defined a port number is selected for initial contact. This is called the well-known port. For example the well-known port for HTTP Web access is 80. When a remote user attempts to connect it sends the request to TCP port 80. Once initial connection is established both computers agree to a different combination of ports to use for ongoing communication. An analogy is to think of well-known port as a doorbell. If ISP blocks access to well-known port remote users are unable to connect.

It is common practice for residential ISPs to block incoming port 80 to prevent customers from running web servers, port 25 used for mail servers to prevent spam, and ports 137, 138, 139, and 445 to prevent remote access to Windows LAN based SMB file sharing. In an effort to reduce file trading some ISPs throttle or block ports used for peer-to-peer (P2P) file trading applications. Impact of blocked ports varies. For example: I have a registered domain. I use outgoing TCP port 25 to communicate with my SMTP mail server to send email. If my ISP blocks outgoing port 25 I will no longer able to send mail through my SMTP mail server.

To get around blocked port it is easy to reconfigure server to use a non-standard port. If access is limited to a small group of friends it is easy enough to simply inform everyone which port to use. If goal is wider public access use of nonstandard ports is a problem. Without knowing port number remote user is unable to connect. URL forwarding is a technique to work around this restriction.

1.5 Traffic Shaping

Internet is as an egalitarian best effort network. This means as packets arrive they are processed on a first come first serve basis. With enough channel capacity incoming packets never have to wait.

Residential ISPs made assumptions about typical customer usage when they set monthly charges and designed infrastructure. Business model assumed bursty data flow predominantly web browsing, email, and occasional file download. Proliferation of Peer-to-Peer (P2P) file trading and streaming video services, such as YouTube and IPTV upset these assumptions. ISPs are struggling to carry more traffic than originally planned.

Some ISPs are responding with traffic quotas. When customer exceeds quota either speed is reduced or additional charge incurred. There have been numerous stories of unwitting customers being billed for thousands of dollars in overage charges on their cell phone data account. One the other hand some ISPs detect undesirable traffic and throttle speed rather than blocking it entirely.

1.6 Deep Packet Inspection (DPI)

Some ISPs use a technique called Deep Packet Inspection (DPI) to determine how customer is using the Internet and block or throttle use they deem harmful. DPI can also be used to obtain additional information about customer’s Internet usage. This data is of interest to targeted marketing vendors. The use of DPI falls into a grey area of what is and is not acceptable ISP behavior. In addition many governments want to know about what their citizens are doing and press ISPs to track customer usage.

1.7 Latency vs Speed

In the quest for ever-faster speed it is important not to lose sight of the interplay between speed and latency. As an example a truck carrying DVDs exhibits very high speed (bits per second) once it arrives but also high latency because it takes hours or days for the data to arrive. Round trop latency is defined as time it takes a packet to go from source to destination and back again. Factors affecting latency are: connection speed, modem overhead, distance, propagation speed, and network congestion.

Modems operate on “chunks” of data increasing latency because entire block must be processed before being passed to next stage. Data cannot be used until the last bit in the bock is received. DSL modems often use a technique called interleave to reduce sensitivity to transient noise. This is effective in maximizing robustness by reducing effect of errors but adds latency because it operates over a larger data block. Low speed connections such as dialup often used smaller packet size to minimize this effect.

Light travels 186,000 miles per second in vacuum. Optical fiber is somewhat slower about 70% of light in vacuum. A packet traveling the 3,000 from New York to LA takes about 25 ms in each direction. To this one must add delay at each router between source and destination. Normally this delay is negligible but if network becomes congested router must temporally store incoming packets until outgoing path is free. In extreme cases router will discard packets. When packets are lost upper level protocol either requests retransmission (TCP/IP) or in the case of streaming data (UDP/IP) fakes missing data.

Impact of latency is heavily dependent on data type. Interactive use such as gaming and Voice over IP (VoIP) telephony place stringent demands on latency but do not require much bandwidth. File transfer on the other hand is relatively insensitive to latency but places great importance on speed.

Typical first-hop latency: T1 or FTTP 1ms, Cable/DSL 10-30ms, Dialup 100 ms, Satellite 500ms. For a more in-depth explanation see “It’s the Latency Stupid.”

1.8 Asymmetric Speed

Most residential broadband service is asymmetric: download is much faster than upload. This is done for technical and business reasons. The business goal is to position residential broadband as primarily consumption based and discourage customer from running servers. Asymmetric speed allows ISP to position residential service differently then business and charge higher fee for business class service.

Low upload speed makes it difficult to run a server or use Voice over IP since upload pipe is easily saturated.

1.9 Measuring Speed

LAN performance is rarely a determinate of Internet speed. Speed is typically limited by first-mile WAN connection. It can be a challenge teasing out various components of end-to-end performance to see if ISP is working as advertised.

IP transmission splits data into 1500 byte chunks called packets (1-byte = 8-bits). Some of the 1500 bytes are used for network control so are not available for user data. TCP/IPv4 uses 40 (TCP/IPv6 60 bytes) of the 1500 bytes for control. NOTE: this analysis assumes use of maximum size packets. Since overhead is fixed using smaller packet incurs higher overhead. With 40-bytes reserved for control out of every 1500-bytes sent only 1460 are available for data. This represents 2.6% overhead.

Some ISPs, typically phone companies, use a protocol called Peer to Peer Protocol over Ethernet (PPPoE) to transport DSL data. This is an adaptation of PPP used by dialup ISPs. Telco’s like PPPoE because it facilitates support of third party ISPs as mandated by the FCC. PPPoE appends 8-bytes to each packet increasing overhead to 48-bytes reducing payload to 1452. Where PPPoE is used overhead is increased to 3.2%.

Many phone companies use IP over Asynchronous Transfer Mode (ATM) (AAL5) to carry DSL traffic. ATM was designed for low latency voice telephony. When used for data it adds significant overhead. ATM transports data in 53-byte Cells of which only 48 are payload the other 5 are control. Each 1500-byte packet is split into multiple ATM cells. A 1500-byte packet requires 32 cells (32 x 48 = 1,536 bytes). The extra 36-bytes are padded, further reducing ATM efficiency. 32 ATM cells require modem transmit 1,696 bytes of which only 1452 carry payload. Where ATM/PPPoE is used overhead is increased to 14.4%.

TCP/IP overhead 2.6% efficiency 97.4%

TCP/IP/PPPoE overhead 3.2% efficiency 96.8%

TCP/IP/PPPoE over ATM overhead 14.4%, efficiency 85.6%

It is easy to determine best-case file transfer rate if modem data rate is known. Broadband marketing rate may not the same as modem transfer rate. This is typically done to simplify marketing by presenting a nice round number. Some Telco’s set transfer rate higher then marketing speed. When customer performs speed test they receive value close to marketed speed. Most broadband modems have status page allowing user to observe true transfer rate. This is rate modem connects to ISP not speed computer connects to modem or router which is typically 10 Mbps, 100 Mbps or 1 Gbps.

As an example our FairPoint 3000/768 ADSL service has a sync rate of 3360/864, 3360 kbps toward customer, 864 kbps toward Internet. FairPoint uses PPPoE and ATM yielding an overhead of 14.4%. Best-case transfer rate is 85.6% of sync rate, resulting in 2,876 kbps down 740 kbps up. Typical file transfer speed reported by Broadband Reports and Speedtest.net are shown below.

NOTE: This is best-case speed. Errors, transmission delays, etc. will reduce speed from this value. The higher the speed the greater the impact of even modest impairments on thru put. Some ISPs offer temporary boost during file transfer that may result in overly optimist report.

1.10 Speed Optimization

There are many urban myths about magical performance improving tweaks. System tuning can be difficult because measurements are hard to duplicate. Many factors are outside user’s direct control.

TCP requires receiver to periodically send an Acknowledge to let sender know everything is OK. If the transmitter has not received acknowledgement after it sends a number of packets it stops transmitting and waits. This is called the receive window. For high speed connection or where latency is high default receive window (RWIN) should be increased to prevent pauses in transmission. Most modern Operating Systems do a good job optimizing RWIN so little is gained by changing it.

If router supports QoS having it give ACKs priority will improve file transfer rate if upload becomes congested.

The other important tweak is packet size, called the maximum transmission unit (MTU). Maximum packet size is typically limited to1500 bytes. Normally this setting is fine for broadband access, dialup uses a much lower MTU typically 576. PPPoE encapsulation adds 8 bytes to each packet. This reduces maximum packet size to 1492 bytes. If sender attempts to transmit a larger packet it will either be rejected or fragmented into two parts, with attendant performance degradation.

A suite of optimization tools is available at Broadband Reports Tools. Once the optimum setting is determined DrTCP utility can used to effect change.

1.11 Servers and Dynamic IP Allocation

Most residential accounts are configured automatically each time customer connects. Dynamically assigned IP address makes it difficult to run a server because address may change at any time preventing remote users from connecting until they learn new address. Dynamic DNS service provides a workaround to run servers on dynamic accounts. A daemon runs on either the router or server to detect address changes. When a change occurs it notifies DNS service which in turn automatically updates A records for the site. Even with automatic DNS update there will still a period of time after the address changes where server is not accessible and active sessions are aborted. Dynamic DNS services are really only suitable for casual personal servers, not business use.

1.12 When “Unlimited” Doesn’t Mean “Unlimited”

There has been much press about residential ISPs marketing unlimited service and then imposing usage caps for heavy users. Some ISPs have gone so far as to call heavy users bandwidth hogs.

Controversy is not over ISP’s right to set terms of use it about misleading marketing. It is about calling a service unlimited then throttling or disconnecting customer if they use it too much.

1.13 When “Always On” Doesn’t Mean “Always On”

Broadband service is marketed as “always on.” Exactly what this means is subject to interpretation. The most “on” service is a bridged or routed connection configured with a static IP address. Once service is configured connection is permanent and always available until the next time the ISP needs to reallocate IP addresses or power fails.

Dynamic Host Configuration Protocol (DHCP) assigns client an IPv4 address for a limited period called a lease. Before lease expires client automatically attempts to renew lease. DHCP simplifies task of managing customer settings. From customer’s perspective service is always on, lease renewal is transparent. Some ISPs bind IP address to hardware MAC address. The same IP address is assigned as long as customer does not change equipment. IPv6 uses a somewhat different mechanism DHCP-PD or Router Advertisement but the end result is the same, customer equipment is automatically configured by the ISP.

Point-to-Point-Protocol over Ethernet (PPPoE) or ATM (PPPoA) emulates traditional PPP dialup connection. This type of service is common for ADSL. It leverages ISP investment in RADIUS authentication and billing equipment. Once customer is authenticated ISP issues IP address. If connection becomes idle the user is disconnected. Most residential routers include a keep-alive mechanism so connection is never disconnected.

Some ISPs limit maximum connect time. After a certain number of hours connection is dropped and must be reestablished. This sort of behavior is common for dialup ISPs and Wi-Fi Hotspots. When connection is dropped customer must log in again to regain Internet access.

1.14 Security and Privacy

Internet is a rough and tumble world often likened to the Wild West. The power of worldwide connectivity means anyone on the planet with an Internet connection is in a position to attack another connected computer. ISPs often block certain ports to reduce danger to unsophisticated users. Port blocking is a double edge sword as it may interfere with customer’s legitimate use of the Internet. Some ISP's go further acting as a firewall protecting customer from hostile attack and examining email for dangerous content or attachments. Some users consider this a great feature in the battle against spam and viruses. Others see it as an unwelcome intrusion in what should be individual control of network access.

The ISP is privy to all traffic that flows through its system. This raises two concerns, nosey ISPs and subpoenas. ISP can easily monitor how customers use the Internet, what sites they go to, what email they send and receive and in some cases even snoop usernames and passwords if they are sent in the clear. Privacy concerns have been exacerbated with expanded government snooping due to war on terrorism. US government asked ISPs to provide information about customer Internet usage without a court order and in most cases ISP complied. Internationally governments are mandating ISPs retain customer traffic information for years.

ISP’s privacy policy determines how customer information is used and protected. It is reasonable to expect ISP to collect and use information for diagnostic purposes and to improve service. However, some ISPs sell or otherwise make use of customer’s browsing data, for example as a way to create targeted ads.

Popularity of wireless networks raises security concerns. In a wired network an attacker must physical connect to the network. With wireless an attacker can eavesdrop from some distance away. This is especially worrisome with Wi-Fi hot spots since they are in public places and integrity of owner is often unknown. To protect against this when setting up Wi-Fi at home use Wireless Protected Access WPA2. This provides over the air encryption. When using public Hot Spots one should be careful accessing any resource over a wireless network where passwords are exchanged in the clear.

IPv6 addressing presents another possible security issue. One of the addressing schemes uses the 48-bit MAC address for the low order bits of the 128-bit IPv4 address. This means hard coded machine MAC address that is not visible outside the LAN in IPv4 becomes part of the pubic IP address and remains the same even when connected to a different ISP. A solution to this problem is to have the computer use a random number rather than the MAC address.

1.15 Network Neutrality

As Internet access becomes pervasive there is growing tension between ISP business practices and public policy. ISPs are concerned about being relegated to commodity bandwidth provider. As such they are frantically trying to create business relationships with select third parties to offer bundled services.

Network Neutrality proponents are concerned ISPs will created walled gardens and be in position to favor some companies and disadvantage others. Opponents of Network Neutrality argue ISPs ought to be able to do anything they want with their own network.

The reason I went into so much detail earlier about required and optional ISP services was to identify those services that only an ISP is able to deliver. Network Neutrality ought to insure network transparency is maintained, innovation encouraged and ISP allowed to offer value add services while being prevented from acting as gatekeeper. Internet’s rapid rise in popularity is the result of its open architecture. Entrepreneurs need to be able to create new business models and interact with customers without requiring permission or cooperation of the network owner.

1.16 Finding an ISP

It can be difficult finding information about ISPs servicing a specific location. For residential use first place to check is local phone and cable companies. Some State Public Utility Commission’s (PUC) maintain broadband provider information.

For WISP checkout Broadband Wireless Exchange Magazine: WISP locator service.

2 Plain Old Telephone Service (Dialup)

Dialup has come a long way from Bell 103 acoustic modem operating at 300 bps to current crop of V.90/92 modems capable of over 50,000 bps. Dialup Internet access is available anywhere there is telephone service. It will even work on cellular at very low speed in a pinch. Almost all Dialup ISPs support ITU-T V.90/92 standard. V.90 modems deliver up to 56 kbps (download) over the PSTN. In the US FCC power limitation reduces effective maximum speed to 53 kbps. V.90 transmission from subscriber to ISP (upload) uses V.34 mode limiting maximum upload speed to 33.6 kbps. If modem cannot connect in V.90 mode it automatically falls back to V.34 mode in both directions with a maximum speed of 33.6 kbps.

V.92 is a minor enhancement to V.90. Upload speed is increased slightly to 48 kbps and implements faster auto negotiation to reduce call setup time. V.44 improves compression of reference test data to 6:1 vs 4:1 with V.90. Compression increases apparent speed because it reduces the number of bits transmitted over slow telephone network. Modem on Hold (MOH) allows modem to park a data session allowing user to answer a short incoming call. This works in conjunction with Phone Company Call Waiting feature and requires support from the ISP.

V.90/92 Dialup Modem End-to-End

V.90/92 requires ISP modem connect to phone company digital trunk. Only a single digital to analog conversion can exist between ISP and user. Phone lines are analog between customer and central office or remote terminal. At that point they are digitized at 64 kbps. This means POTS modem technology has reached its theoretical maximum speed. To obtain higher speed requires use of different technology.

At connect time modem probes phone line to determine noise and attenuation characteristics in order to set initial connect speed. Speed is constantly adjusted in response to varying line conditions. To obtain maximum speed V.90 and V. 92 modems require phone circuit that exceeds minimum FCC requirements.

2.1 Dial Up Networking (DUN)

Dialup networking (DUN) is used to establish an Internet connection. The most common method used to traverse the telephone network is via Point-to-Point Protocol (PPP). PPP allows Internet Protocol (IP) packets to traverse the serial point-to-point telephone link between user and ISP. DUN automatically dials ISP phone number, waits for modem to connect and establishes PPP session. The ISP performs user authentication and assigns an IP address. DUN monitors the connection and notifies user when it disconnects. In Windows Internet Explorer can automatically activate DUN when attempting to connect to a web site.

2.2 Session Duration

Dialup ISP business model assumes customer stay connected for relatively short periods of time. To enforce this most ISP’s automatically disconnect customer when limit is reached. Session will also be dropped due to extended inactivity.

2.3 Multilink

In the quest for higher speed some dialup ISPs support Multilink. Multilink binds two dialup links into a single faster connection. If customer typically connects at say 44 kbps multilink doubles speed to 88 kbps. Multilink requires two modems; two phone lines, and an ISP that supports it. Where available it is a useful technique to obtain better performance from dialup.

Software at each end of the link splits data between each connection effectively doubling speed. Unfortunately because data is still traveling over low speed dialup multilink does not improve latency.

Multilink is also used with ISDN to bind the two bearer channels together for 128 kbps connection.

2.4 Impairments

2.4.1 Slow Speed

Modem data connection is more demanding then voice. There are many reasons for slow dialup even though phone sounds normal. Dialup modem impairments are discussed at length in a separate paper.

2.4.2 Call Waiting

Call waiting can be temporally disabled at the beginning of a call. The sequence varies by locale, in our area it is *70. Unfortunately sending the disable sequence to a line not equipped with call waiting is interpreted as part of the dialed number, resulting in an incorrect connection. This is a problem if the modem uses multiple lines and not all are equipped with Call Waiting.

2.4.3 Shared Line

If dialup modem shares a phone line with telephone or fax machine there is possibility of mutual interference. If modem is in use picking up a phone will cause modem to disconnect. Conversely if phone is in use modem may attempt to connect interfering with call. One can use a privacy device that monitors phone line voltage to prevent this. When phone is idle open circuit voltage is high around 48 volts, when a phone/modem is in use voltage drops to less than 10 volts. Privacy adapters measure line voltage to prevent phone use if a call is already in place. There are a couple of inconvenient side effects to this approach. Privacy device prevents calls being transferred from one phone to another and it confuses line use indicators built into many phones. I designed a Modem Access Adapter to prevent interference when modem and phone share the same line.

3 T-1 and E-1 Carrier

The US Bell System developed T-1 digital carrier during the early 60’s to reduce interoffice transmission cost. Prior to T-1 analog frequency division multiplexing (FDM) was used to carry voice traffic between telephone switching centers. FDM carrier used a 4-wire circuit to carry 24 voice channels, one pair in each direction. T-1 was designed to carry 24 voice channels, facilitating transition from FDM to TDM. E-1 digital carrier, used in Europe, is similar transporting 30 voice channels. Each voice channel is digitized resulting in a 64 kbps data rate. 24 channels require 1.536 Mbps plus an 8 kbps control channel resulting in data rate of 1.544 Mbps (E1 is 2.048 Mbps). T-1 has a DS-1 channel speed of 1.544 Mbps and is carried over a 4-wire copper facility. Popular usage has corrupted this distinction. T-1 is now commonly used to mean any 1.544 Mbps service.

In the early 1980’s T-1 was tariffed and made available to customers. T-1 continues to be extremely popular in commercial service carrying both voice and data. Prices for T-1 have dropped dramatically as technology improves and competitive pressure by alternative high-speed services.

3.1 Converting Voice to Digital Bits

Voice grade phone service occupies the frequency band of 300-3000 Hz. Low frequencies are suppressed to minimize interference from 50/60 Hz power lines. Increasing upper frequency beyond 3000 Hz does little to improve intelligibility, at the expense of greater bandwidth. Digital sampling must be performed at least twice the highest frequency of interest to recover the original analog signal. Engineers chose a sample rate of 8,000 times a second. It was found sampling to 12-bits, resulting in 4096 possible values, produced excellent voice quality. This required 96 kbps per channel resulting in a composite data rate that exceeded what 1960s technology could deliver. To reduce data rate engineers decided to use only 8-bits or 256 values per sample, resulting in a 64 kbps data stream. To minimize quality degradation, conversion is performed logarithmically. When sound level is low samples are close together. During loud passages samples are farther apart. This masks quantizing noise generated by the conversion process. Two slightly different methods are used, u-LAW in US and A-LAW in Europe. The resulting digital signal is called Pulse Code Modulation (PCM). 24 phone calls in US (T-1) or 30 Europe (E-1) are interleaved using Time Division Multiplexing (TDM) combined with an 8 kbps signaling channel the composite data stream is 1.544 Mbps (US) or 2.048 Mbps Europe.

PCM coding scheme developed for T-1 is what makes V.90 and V.92 dialup modems possible and also the reason dialup is limited to 56 kbps. Logarithmic sampling minimizes effect of audible noise but only allows 7 of the 8 bits be used for data. 8,000 samples per second times 7-bits per sample results in maximum data rate of 56,000 bits per second. Dialup modems have reached their theoretical limit.

3.2 Channelized vs. Unchannelized

When used for Internet access voice channelization is neither required nor desired. T-1 data circuits are unchannelized exposing total channel capacity to the IP layer. IP, rather than T-1, performs multiplexing. Some circuits are provisioned to allow flexible control of channelization. This allows an Integrated Access Device (IAD) to dynamically allocate bandwidth between voice and data.

3.3 Provisioning

The original implementation of T-1 required regenerators spaced every 6,000 feet. Regenerators recreate bipolar signals, allowing T-1 to deliver very low error rates compared to analog carrier. Regenerators can be powered from the T-1 line, called a span, eliminating need for local power. T-1 bipolar signaling is relatively noisy. This requires care during circuit provisioning to prevent interference between T-1 and other services, including other T-1s and DSL in the same cable.

Early T1 required a 4-wire circuit, 1-pair in each direction. Newer T1 deployments using HDSL2 only need a single pair. Digital signal processing techniques similar to that used with DSL reduce outside plant cable requirement and increases distance between regenerators.

4-wire T-1 circuit can be up to 50 miles, with regenerator every mile. Very long T-1 circuits are rare nowadays as fiber is more cost effective.

3.4 CSU and DSU

Channel Service Unit (CSU) is connected directly to the 4-wire facility. The CSU regenerates T-1 bipolar signals before handing them off to Data Service Unit (DSU). The CSU provides keep alive and Loopback testing enabling Telco to monitor line quality.

T-1 uses bipolar plus and minus 3-volt pulses, between pulses line voltage returns to zero. The Digital Service Unit (DSU) converts bipolar signals to a synchronous interface such as V.35 using both RS232 single ended and RS422 differential signaling to connect to customer equipment.

In the US CSU and DSU are built into customer premise equipment (CPE), such as a T-1 router. In the rest of the world CSU is owned by service provider, CPE includes only the DSU.

3.5 Smart Jack

When T-1 was developed the interface between CSU and DSU, called DSX-1, was designated the demarcation point between Telco and customer. DSX-1 is still the demarcation point in rest of the world. During US deregulation the FCC defined the 4-wire facility as the demarcation point. This caused problems for service providers as now management and quality assurance functions were no longer under their control but provided by customer premise equipment (CPE).

The solution was the Smart Jack. It presents a 4-wire (2-pair) interface to customer and implements service provider Loopback test function. This allows Telco to perform testing and maintenance functions while complying with FCC regulations.

Smart Jacks can be also used to deliver T-1 service to customers by converting other transmission schemes to traditional 2-pair T-1.

3.6 Beyond T-1

The modern telephone network is almost entirely digital except for 2-wire analog POTS. As technology improves many voice calls can be carried over a single circuit. The digital carrier hierarchy is based on the notion of voice channels. The lowest level, called Digital Service 0 (DS-0), is a single PCM digitized voice circuit of 64 kbps. Next in the hierarchy is DS-1 (24 voice circuits over T-1 carrier) operating at 1.544 Mbps, then DS-2 (T-2) operating at 6.312 Mbps equivalent to 4 T-1 circuits. Then DS-3 (T-3) at 44.736 Mbps equivalent to 28 T-1 circuits.

Higher speed transmission is optical using Synchronous Optical Network (SONET) and ITU Synchronous Digital Hierarchy (SDH). Optical Carrier 1 (OC-1) and Synchronous Transport Signal Level 1 (STS-1) operate at 51.84 Mbps. Next in the hierarchy is STS-3 (OC-3) 155.52 Mbps. Then STS-12 (OC-12) operating at 622.08 Mbps and so forth. 10 Gbps STS-192 (OC-192) is an interesting convergence point. It is the first time Ethernet and SONET/STS operate at the same speed opening the door for Ethernet being carried directly over SONET.

For years 1.544 Mbps T-1 service was considered fast, and compared to dialup it is. Advent of low cost residential broadband is driving demand for ever faster commercial service at the same time it is exerting downward price pressure. The good news for carriers is the tremendous capacity of fiber. Once fiber is installed increasing speed is often just a matter of changing access setting – cable plant and electronics are not affected. Even when electronics needs to be upgraded existing cable plant remains unchanged.

Telco Digital Carrier Hierarchy
301.336 Gbps	OC-7144F
159.252 Gbps	OC-3072 STS-3072
100.000 Gbps		100 Gig Ethernet (not part of digital hierarchy)
79.626 Gbps	OC-1536
39.812 Gbps	OC-768 STS-768	40G Sonet Telco convergence with 40 Gig Ethernet
9.953 Gbps	OC-192 STS-192	10G Sonet Telco convergence with 10 Gig Ethernet
2.488 Gbps	OC-48 STS-48
1.244 Gbps	OC-24 STS-24
1.000 Gbps		Gig Ethernet (not part of digital hierarchy)
622.080 Mbps	OC-12 STS-12
155.520 Mbps	OC-3 STS-3
100.000 Mbps		Fast Ethernet (not part of digital hierarchy)
51.840 Mbps	OC-1 STS-1
44.736 Mbps	T-3 DS-3 North America
34.368 Mbps	E-3 Europe
10.000 Mbps		Ethernet (not part of digital hierarchy)
8.448 Mbps	E-2 Europe
6.312 Mbps	T-2 DS-2 North America
2.048 Mbps	E-1 DS-1 Europe	30 DS-0 voice channels
1.544 Mbps	T-1 DS-1 North America	24 DS-0 voice channels, Primary Rate ISDN
144 kbps	Basic Rate ISDN	2B (64 kbps bearer channel) + D (16 kbps data channel)
64 kbps	DS-0	(DS) Digital Signal, Single PCM voice channel

4 Integrated Service Digital Network (ISDN)

Success of T1/E1 prompted Telephone industry to look for a way to deliver high-speed digital service directly to customer. Integrated Service Digital Network (ISDN) was supposed to be the next big thing poised to revolutionize telephone industry. Things have not played out that way. Deployment missteps and high cost have slowed deployment. ISDN is viable where other forms of high-speed access are not unavailable but its window of opportunity has passed.

Basic rate ISDN provides two 64 kbps bearer channels (B channels), and a 16 kbps data control channel (D channel). Primary Rate ISDN is basically a T-1 connection. ISDN is a circuit switched technology with very fast call setup time. Being digital full 64 kbps is available.

ISDN requires a Terminal Adapter (TA). TA connects to ISDN line, provides two POTS analog phone lines, and serial data connection. Both B channels may be bonded into a single faster 128 kbps connection. Combined with automatic dialing and incoming call detection both channels can be used for maximum speed when needed, then automatically falling back to single channel during voice call. ISP needs to support multilink in order to bond both B channels to provide 128 kbps service.

ISDN, like T1/E1, allows use of regenerators to extend distance between customer and central office.

Cost Tip – ISDN is a switched service. Make sure ISP has access numbers, Points of Presence (POP); close enough so calls are flat rate. Failure to do so will result in a rude surprise when the phone bill arrives. Telco’s offer different types of ISDN service for Internet access unmetered is ideal.

4.1 IDSL

A variation of ISDN, ISDN digital subscriber line (IDSL), uses ISDN signaling to deliver 144 kbps data only service at greater distance than typical DSL.

5 Digital Subscriber Line (xDSL)

Digital Subscriber Line (DSL) technology utilizes telephone copper wiring between subscriber and phone company central office (CO) or Remote Terminal (RT) to deliver high-speed data. This allows local exchange carrier (LEC) to generate additional revenue by leveraging its massive investment in copper outside plant cabling. Several types of DSL have been developed hence the xDSL moniker. The most common types are Asymmetric DSL (ADSL) G992.1, ADSL2 (G.992.3), ADSL2+ (G.9925) and Symmetric DSL (SDSL). Telco’s like DSL not only as another revenue source but because it gets long duration data calls off the Public Switched Telephone Network (PSTN). This minimizes need for expensive upgrades to circuit switched phone network.

ADSL was initially developed for video on demand and has been recycled for broadband Internet access. ADSL has higher download speed, toward the subscriber, than upload. It uses frequencies above those used with Plain Old Telephone Service (POTS) allowing it to coexist with voice service. This minimizes cost by allowing a single copper pair to be used for both voice and data. A customer can be equipped with both phone and high-speed data service over the same copper circuit. Typical ADSL speed is 768 – 8,000 kbps downstream (toward customer) and 128 - 800 kbps upstream (toward Internet). ADSL2+ doubles maximum download rate over short loops.

The Digital Subscriber Line Access Multiplexer (DSLAM) at the Telephone Central Office or Remote Terminal is connected to the customer’s phone line. The voice portion is passed through a low pass filter and delivered to POTS phone switch. DSLAM recovers customer data and uses Asynchronous Transfer Mode (ATM) to link customer to ISP. Telco’s use ATM because it facilitates support of third party ISPs. At the customer location a similar filter is used to separate DSL from POTS. This can be a single whole house POTS/DSL splitter or filter connected ahead of each non-DSL device.

Symmetric DSL (SDSL) is typically marketed as a business service. It requires a dedicated copper pair not shared with POTS. Being symmetric makes it suitable for use with servers.

Maximum DSL speed is a function of line length, wire gauge and line quality. ADSL service is limited to about 18,000 feet, with closer customers eligible for higher speed. Remote DSLAMs, called Remote Terminals (RT), shorten loop distance by moving the DSLAM closer to the customer. This increases number of potential customers within range and increases maximum speed.

Aware published several White Papers reviewing different versions of ADSL.

Figure 14 ADSL First-Mile

FCC regulations require Incumbent Local Exchange Carrier (ILEC) allow third party data local exchange carriers (DLEC) access to copper access network. Copper subscriber loop is tariffed as an unbundled network element (UNE). DLECs rent collocation space within the central office and install their own DSLAMs and backhaul facilities. Even ILECs need to set up a separate entity to deliver DSL because unlike phone service data is not a regulated service.

DSL can also be configured as a wholesale service. ISP enters into an agreement with a DLEC. The DLEC in turn rents copper circuit from the ILEC, installs DSLAM within the central office and transports customer traffic to the ISP’s point of presence. This is why Telco’s like using ATM to deliver DSL It allows them to offer wholesale DSL service. As such they evolved a complex interconnect scheme consisting of 1) physical copper loop, 2) Asynchronous Transfer Mode (ATM) virtual circuits to transport customer packets over the First-Mile network to the respective ISP 3) typical ISP functions. DSL may involve three different companies, ILEC supplying copper service, CLEC first mile transport, and ISP doing the rest.

Wholesale DSL service has not worked out well; most of the players have left the market. Colocation is still viable for copper. The FCC has not deemed fiber an unbundled network element so DLECs are often not able to rent dark fiber, limited to copper. Covad is probably the best known DLEC.

5.1 Splitter vs Inline Filter

ADSL and Voice telephone share a single copper circuit. At each end filters prevent DSL from interfering with voice phone service. To reduce cost ADSL service providers include inline filters as part of a customer self-install kit. Customer is instructed to installed filter at each non-DSL device. Having customer self-install filters eliminates expense of a truck roll.

An alternative to per device filtering is a whole house POTS/DSL splitter. Splitter provides low pass filter isolating voice from high frequency DSL. Splitter has two outputs; “Data” connected directly to DSL modem and “Voice” connected to inside phone wiring. Some splitters contain a half-ringer test circuit after the low pass POTS filter. This allows removal of half-ringer in NID, minimizing DSL signal loading.

Inline Filter and Whole House POTS/DSL Splitter

Splitter Advantage

Splitter Disadvantage

Single device for entire house
Better electrical characteristics
Isolates inside wiring from DSL
Isolates half-ringer from DSL
Works with home Alarm dialer

Installation required
Dedicated run from splitter to DSL modem
Have to purchase separately

5.2 Impairments

DSL uses 100-year old copper telephone network to carry high-speed data. This is an impressive engineering accomplishment. Unfortunately not all phone lines are suitable for DSL. Assuming the local central office (CO) or remote terminal (RT) is equipped for DSL it may not be available for a number of reasons. This section discusses common problems and where applicable workarounds.

5.2.1 Loop Carrier

Digital Loop Carrier (DLC), Digital Added Main Line (DAML) et al are techniques that allow multiple telephone customers to share a single copper circuit. Phone companies use loop carrier when there are no available circuits and in rural areas where cost of active electronics is less then running a dedicated circuit to customer. Unfortunately most forms of DLC are incompatible with DSL.

5.2.2 Network Interface Device (NID)

In the bad old days before US telecom divestiture (1880’s to early 1980’s) Phone Company supplied service, owned customer premise equipment (CPE) and leased it to customer. Customer was prohibited from connecting anything to the telephone network. With divestiture Phone Company regulated responsibility was limited to delivering service to premise. Inside wiring and CPE became the customer’s responsibility.

This created a dilemma for the Phone Company, how to determine if a problem was their responsibility or that of the customer? Enter the Network Interface Device (NID). NID is the demarcation point, between Phone Company and customer. It incorporates lightning protection and a method to easily disconnect customer premise equipment (CPE) from the telephone network. Early NIDs used modular jack connected to old style carbon block lightning protector. Over time NIDs evolved into an integrated package and gas tube surge protection replaced carbon block. Gas tube protection is preferred because module is hermetically sealed, provides more consistent over voltage protection and has lower shunt capacitance then carbon block. Carbon protectors have to tendency to increase circuit noise over time. This may cause problems if DSL signal is weak.

Phone Company use automatic test equipment called mechanized loop test (MLT) to periodically test copper phone lines. They wanted a device; built into the NID, which allowed MLT to determine where the network ended and where customer responsibility began. There have been two different approaches to this: MTU and Half-Ringer.

5.2.2.1 Maintenance Terminating Unit (MTU)

MTU was developed during the early days of deregulation. It is a pretty clever device; it consists of a series pass voltage controlled switch on each leg of the Plain Old Telephone Service (POTS) circuit. During normal phone usage switch conducts and POTS equipment operate normally. Testing, done at a lower voltage, does not trigger the series element thus isolating CPE side from the telephone network. MTU being a series pass device has four leads, two connect to Telco side the other two to CPE.

Unfortunately MTU’s are incompatible with DSL. DSL modem does not draw current from phone line so MTU never turns on. The MTU isolates DSL modem from the telephone network.

If your line has an MTU it must be removed. For reasons too involved to go into here MTU’s caused other problems and have not been used for years. If your phone line had an MTU it was most likely removed years ago, but it is possible some were missed. Automated testing should flag the existence of an MTU, but not always.

5.2.2.2 Half-Ringer

Half-Ringer is a simple circuit that emulates old style electromechanical Western Electric ringer providing a test signature for automatic test equipment. It consists of a capacitor, resistor, and back-to-back zener diodes. ADSL is designed to operate in the presence of a half ringer so in most cases it will have no effect on ADSL. It does represent a small load so if signal is marginal removing it may help. SDSL is not able to operate in the presence of a Half-Ringer and it must be removed.

Excerpt from Broadband Forum Technical Report 013.

It has been standard practice in many areas of the United States to install, at the Network Interface Device (NID), a network termination device called a half ringer. It is an example of an AC termination device since it is detected using AC techniques.

A normal POTS mechanical ringer is made up of an inductor and capacitor in series bridged between Tip and Ring. The half ringer is a capacitor in series with a zener diode and resistor. This, in the U.S., is a 0.47 micro Farad capacitor without the addition of the inductive part of the circuit, hence the name ‘half’ ringer.

5.2.3 Distance

ADSL service is limited to about 18,000 feet. Some ILECs are installing Remote Terminals (RT) to reduce cable distance allowing them to serve more customers at higher speed. Newer DSL equipment has slightly extended maximum distance. Obtaining accurate pre order distance estimate can be a difficult. The effective wire distance between DSLAM and customer is often substantially longer than driving distance making map based estimates questionable.

5.2.4 Bridged Taps

When telephone feeder cable is installed it is not known how many customer circuits will be needed at each location. The solution is to run a large feeder cable past many customers. As service is installed the technician selects an unused cable pair and splices it to the drop cable. The circuit feeding the drop may continue for hundreds or thousands of feet beyond the customer, causing a bridged tap. Bridged taps are of no consequence for telephone service but can degrade DSL. The presence of a bridged tap causes DSL signal split at the tap going down both paths. When it reaches the end it is reflected back into the line, creating interference. DSL is designed to tolerate some amount of bridged tap, but if circuit is marginal it may cause problems or push customer over distance limit. SDSL providers typically pay Telco to remove bridged taps. This is expensive and is not ordinarily done for low cost residential ADSL.

5.2.5 Noise and Crosstalk

Telco feeder cable carries many different services: POTS, ISDN, DSL and T-1. Phone circuits often closely parallel power lines picking up power line noise. Imperfections cause unintentional coupling from one circuit to another. This raises the noise floor. If noise becomes excessive speed is impacted.

Residential DSL is not typically warranted for speed, it is a best effort service. If noise is present during phone call customer is more likely to get problem resolved than if it only affects DSL or dialup.

5.2.6 Load Coils

Resistance and impedance attenuate signals. This effect is more pronounced at high frequencies and long circuits. Load coils are used on long loops to cancel these harmful effects resulting in better voice characteristics. Load coils are typically installed on loops over 18,000 feet. H88 load coils, the most common type, are spaced every 6,000 feet beginning 3,000 feet from the central office.

Unfortunately load coils are incompatible with DSL. They flatten response over the voice frequency range but severely attenuate higher frequencies used by DSL. If Load coils are present they must be removed prior to getting DSL.

5.2.7 Interleave vs Fastpath

DSL lines are susceptible to noise bursts from many sources such as: lightning, ignition noise, radio transmissions and power line faults. DSL spec writers were aware of this and included forward error correction (FEC). FEC adds redundant check bits to data stream. If noise corrupts data check bits are used to recover from error. As long as only a few bits are damaged receiver is able to correct errors avoiding need for retransmission.

If noise burst is long it corrupts too many bits for receiver to undo damage. In that case bad packet is passed to higher layer protocol. TCP/IP TCP requests retransmission. Needless to say that takes a "long" time. UDP/IP, used with VoIP and streaming data does not have a retransmission scheme. There is not enough time to retransmit data before it is needed by application. Streaming applications have provisions to fake missing data. How bad missing data affects quality depends on the application and how extensive the damaged.

When interleave is turned on bits from several frames are interleaved in time. If noise burst is long relative to bit time it corrupts many bits. When receiver deinterleaves data corrupt bits are now spread across multiple frames - increasing chance FEC is able to correct them. This eliminates need for retransmission or application having to fake missing data.

As speed increases number of bits affected by a given noise burst increases. Let’s say a noise burst corrup