Chapter 20. Internet Service Providers

Once upon a time end users would use a computer and a modem to dial a provider. The provider would have a modem connected to a server and this would allow simple text access to a computer system. This was then used to provide bulletin boards.

This moved on, and providers started to allow direct Internet Protocol (IP) access to end users. The modem would connect and the computer would authenticate and pass IP packets using protocols such as SLIP and PPP. This allowed the computer to authenticate and be allocated an IP address.

Point to Point Protocol (PPP) worked well and is still in use today in broadband access networks. The modem at the provider would connect to the provider's network and the Internet. Typically there would be one device, an Access Concentrator which connected IP on one side and modems on the other. The IPs would be fixed for each modem (so dynamic for the end user as depends which port they hit) and routing could be static to each Access Concentrator.

PPP is quite a simple protocol that allows packets to be marked with their type, but it also provides negotiation protocols for Link Control (LCP), authentication (CHAP and PAP), and IP level negotiations (IPCP and IPV6CP). Once negotiation is complete then IP packets can be passed using PPP.

As networks became more complex a separation of the Access Concentrator into a L2TP Access Concentrator (LAC) which has the modems, and the L2TP Network Server (LNS) was sensible. The LAC accepted the call on the modem and established a Layer 2 Tunnelling Protocol (L2TP) connection to the LNS. This allowed PPP to be passed from the end user computer to the LNS. The LNS is responsible for the PPP negotiations and passing IP packets to and from the Internet.

L2TP provides a simple means for PPP packets to be passed over an IP network. It uses a small header and UDP to pass packets between the LAC and LNS.

Sometimes it because sensible for the LAC to decide to which LNS it should connect by some means. A good example is where a carrier with LACs will route connections to wholesale customers' LNSs. This would allow ISPs to make use of providers that have modems. This is actually the way it works on broadband access networks. For example, BT, O2, and TalkTalk have LACs in their network which pass L2TP to their ISP customers.

To achieve this, the LAC does some of the initial PPP negotiations. It handles the LCP and starts the authentication. It then establishes the L2TP connection passing these proxy details on to the LNS. The choice of LNS is done using the username, which is why it has to start the authentication. Typically a realm is included in the user name, using an @ and a string at the end of the username to steer the connection to the right LNS.

In a typical broadband network we don't have dialup modems in the same sense. The modems are jumpered to the phone line at the exchange and are part of an Access Node, usually called a DSLAM or MSAN. This then passes PPP packets on to a Remote Access Server, usually called a BRAS. The link from DSLAM to BRAS is typically PPPoE. The BRAS acts as the LAC and connects to an ISP's LNSs.

PPPoE is PPP over Ethernet. Some access networks use DSL to carry PPP packets directly (PPPoA), and some use the ADSL as an Ethernet Bridge (PPPoE). There are access networks which provide Ethernet by some means to the end user equipment which then communicates via PPPoE to the BRAS. All of these work in much the same way at the BRAS as it sees PPPoE connections.

Typically the BRAS provides the initial proxy negotiation and then establishes an L2TP connection, after which it is no longer involved in any negotiation, but just passes on PPP packets each way.

Remote Access Dial Up Server is a system that allows the authentication decisions and allocation of IP addresses to be passed on to separate servers rather than being configured into the various equipment. RADIUS uses UDP to send a request to a server and send a reply back.

RADIUS is used within carrier networks so that the BRAS can check to where it is to send an L2TP connection. The RADIUS response can contain the tunnel details it needs, including the authentication within L2TP.

RADIUS is also used between a carrier and an ISP. The carrier will send a RADIUS request to the ISP asking the ISP for details of the LNS to which the connection is to be sent. This allows the ISP to steer sessions as they need.

Once the LNS gets the L2TP connection, RADIUS is used to obtain the IP address details to be assigned to the specific connection.

RADIUS is also used for accounting, to provide details of connections in progress and volumes of data transferred.

Appendix G provides details of the specific AVPs used with RADIUS for L2TP.

Once a connection is made to an LNS, the end user is assigned IP addresses. Obviously there is a need to ensure that the IP addresses are routed within the ISPs network to the correct LNS. OSPF and BGP are the main routing protocols used for this (though, back in dialup days, RIP and RIP2 were often used, and were a bit slow). OSPF is not ideal for this as it means the whole OSPF network tracking every connection of every user. The FireBrick supports use of BGP to announce connected IP addresses into an ISPs internal network as connections are made via L2TP.