How Do I Choose the Best Cheap ISP?

The challenge for Internet service providers is mainly how to satisfy their customers and maintain rapid growth.

In the early 1990s, mapping business flows to the physical topology of the network was not achieved in a scientific way. The implementation of this mapping is just a product-based routing configuration-service flows are simply assigned to the shortest path calculated by the Internal Gateway Protocol (IGP) used by the ISP. The limitation of this irregular mapping is solved by providing excessive bandwidth when a link is blocked. Nowadays, ISP networks are getting larger and larger.

The task of mapping business flows onto existing physical topologies is called traffic engineering.

Due to the unprecedented growth in customer demand for network resources, the nature of important tasks in IP applications, and the increasing competition in the Internet market, traffic engineering has become an important issue within ISPs. Existing IGPs do not take bandwidth availability and service characteristics into account when setting up the forwarding table, which can cause network congestion. ISPs understand that traffic engineering can effectively enhance the operation and performance of the network. They want traffic engineering to:

1. When routing the main path, bypass known bottlenecks and blocking points in the network.

2. When one or more failures occur on the main path, provide clear control for how the service reroutes.
3. By ensuring that the auxiliary equipment of the network will not be overused, and at the same time, the auxiliary equipment of the network on the potential alternative path will not be underused, thereby effectively utilizing the available integrated bandwidth and long-distance fiber.
4. Make ISP more competitive in the market by maximizing operational effectiveness and minimizing other operating costs.
5. Enhance service-oriented performance characteristics in the network by minimizing packet loss, minimizing blocking hold time, and maximizing throughput.
6. Enhance the statistically constrained performance characteristics (such as loss rate, delay variation, transmission delay, etc.) of the network to support the multi-service Internet in the future.
7, to provide customers with more choices, lower costs and better service.

In early router-based core networks, traffic engineering was implemented by simply

To implement router-based traffic engineering solutions, Juniper Networks has been actively participating in the formulation of the Multiprotocol Label Switching (MPLS) standard and related IETF working groups. We believe that the strategy for traffic engineering using MPLS includes four basic components:

1. Packet forwarding 2. Information release 3. Path selection 4. Signaling

Each functional unit is an independent module. Juniper Networks' traffic engineering architecture provides an open interface between the four functional modules. This modular combination and open interface provides flexibility for individual modules to make changes as needed when better solutions emerge.

The packet forwarding unit in the Juniper network traffic engineering structure

People generally believe that MPLS can significantly enhance the forwarding performance of LSR. Rather, exact lookups, such as those provided by MPLS and ATM switches, are faster than the longest match lookups provided by IP routers. However, recent chip technology advances have enabled ASIC-based routing query engines to run at the same speed as MPLS or ATM VPI / VCI lookup engines.

The real advantage of MPLS technology is that it provides complete separation between routing (ie, control) and forwarding (ie, transferring data). This separation allows multiple services and service types to be configured using only a single forwarding algorithm, MPLS. In the future, when ISPs need to develop a new value-added service, the MPLS forwarding structure can be retained, and new services can be simply established by replacing the method in which packets are distributed to the LSP. For example, when a packet is assigned to an LSP, it can be based on a combination of destination subnet and application type, a combination of source and destination subnets, special QoS requirements, IP multicast groups, or virtual private network (VPN) identification numbers . Based on this approach, new services can simply be added to the commonly used MPLS forwarding structure.

Because traffic engineering requires detailed information about network topology and network load dynamics, the main requirement of the new traffic engineering model is a framework for information release. This part can be achieved simply by defining the relevant IGP extensions, so that the link characteristics can be included in the link state broadcast of each router. IS-IS extension can be realized by defining a new type length value (TLV), while OSPF extension can be realized by opaque LSA. The standard spreading algorithm used by the link state IGP guarantees that the link characteristics are advertised to all routers in the ISP routing domain.

Each LSR manages network link characteristics and topology information through a special traffic engineering database (TED). TED is used to calculate the external path of the LSP through the physical topology. A separate database is maintained to make concurrent traffic engineering calculations independent of the IGP and IGP link state databases. At the same time, IGP continues to operate unchanged, performing traditional shortest path calculations based on the information contained in the router's link state database.

Some traffic engineering extensions that need to be added to the IGP link state broadcast include:

1. Maximum link bandwidth 2. Shortest reserved bandwidth 3. Current

After the network link characteristics and topology information are diffused through the IGP and stored in the TED, each starting LSR uses the TED to calculate a set of LSP paths that pass through the routing domain. The path of each LSP can be represented as an exact or loose external route. An external route is pre-set by a series of LSRs that are part of the LSP physical path. If the input LSR determines all LSRs in the LSP, the LSP is considered to be determined by precise external routing. If the starting LSR specifies only a few LSRs in the LSP, the LSP is described by loose external routes. Supporting precise and loose external routing at the same time allows the routing process to give both maximum freedom when possible and constraints when needed.

The starting LSR determines the physical path of each LSP by using the Constrained Shortest Path First (CSPF) algorithm on the information in the TED. CSPF is an improved shortest path first algorithm. It is an algorithm that takes certain constraints into account when calculating the shortest path through the network. The inputs of the CSPF algorithm include:

Topology link state information obtained from IGP and maintained in TED.
Properties related to the state of network resources (such as total link bandwidth, scheduled link bandwidth, available link bandwidth, and link color) carried by the IGP extension and stored in the TED.
Obtained from the user settings to support the management features (such as bandwidth requirements, maximum number of hops, and management strategy requirements) required when the service passes the proposed LSP.

When CSPF considers each candidate node and link of a new LSP, it can accept or reject specific path components based on the availability of resources or whether the selected part violates user policy constraints. The output of the CSPF calculation is an extrinsic route, which contains a set of shortest paths through the network and meets the constraints of the LSR address. This external route is then passed to the signaling part, and the LSR of the signaling part in the LSP establishes the forwarding state. The CSPF algorithm is repeated in the initial LSR required to occur within each LSP.

Although online path calculations reduce administrative work, offline planning and analysis tools are needed to optimize global traffic engineering. Online calculations take resource constraints into account and calculate one LSP at a time. The challenge of this implementation is its certainty. The order of LSP calculation will play an important role in determining the physical path of the LSP through the network. The earlier calculated LSP has more effective resources than the later calculated LSP, because the previously calculated LSP consumes network resources. If the order of LSP calculation is changed, the physical path structure of the LSP will also change accordingly.

Offline planning and analysis tools simultaneously examine the resource constraints of each link and the requirements of each input-output LSP. Offline implementation may take several hours to complete. It provides global calculations, compares the results of each calculation, and then selects a global best solution for the network. The output of the offline calculation is a series of optimized LSPs used by network resources. After the offline calculation is complete, LSPs can be established in any order, because all installations of LSPs follow the rules of the global optimization scheme.

Because the information about the network status resident in the TED of the initial LSR is out of date at any time, the path calculated by CSPF is only considered acceptable. Only after the LSP is actually established by the signaling part can it be known whether this path is really working. The signaling responsible for establishing LSP status and bid allocation depends on some extensions of the Resource Reservation Protocol (RSVP):

External routing objects allow RSVP path (PATH) information to be transmitted in an external LSR sequence that is independent of the traditional shortest path IP routing.
The bid request object allows the RSVP path (PATH) information to request the intermediate LSR to provide a bid bundle for LSP establishment.
The bidding object allows RSVP to support the allocation of bidding without changing the existing mechanism. Because the RESV information of RSVP follows the predetermined path of the RSVP path information, the bidding object supports bidding allocation from the downlink node to the uplink node.

RSVP is ideal as a signaling protocol for establishing LSPs:

RSVP is a standard Internet resource reservation protocol.

The essence of traffic engineering is to map the service flow to the physical topology. This means that the core of providing traffic engineering through MPLS is to determine the physical path for each LSP. This path can be determined by offline settings or by online constraint-based routing. Independent of the calculation method of the physical path, the forwarding status can be installed in the network through the RSVP signaling function.

The Juniper network's MPLS-based traffic engineering strategy supports different routing and setting methods for LSPs:

The ISP can perform full path calculation for the LSP offline, and individually set each LSR in the LSP with the required static forwarding state. This is similar to the current setting of IP-over-ATM by some ISPs.
The ISP can perform full path calculation of the LSP offline and perform static full path configuration of the starting LSR. The initial LSR uses RSVP as a dynamic signaling protocol to install a forwarding state for each LSR in the LSP.
ISPs can rely on constraint-based routing to provide online dynamic calculations for LSPs. In constraint-based routing,

To provide a powerful and user-friendly tool, Juniper Networks' traffic engineering structure is designed to support a wide range of customer needs. This implementation allows the network operator to need:

Provides LSPs with many special operations that are important for traffic engineering:
Establish an LSP.
Activate an LSP to start forwarding traffic.
Terminate an LSP to stop forwarding services.
Modify LSP attributes (such as bandwidth, hop limit, and CoS) to manage its performance characteristics.
Reroute the LSP to change the physical path through the network.
Dismantling an LSP causes the network to reclaim all resources allocated to the LSP.
Configure loose or precise explicit routing for LSPs. Support for this choice allows the path selection process to obtain a large degree of freedom where possible or to constrain it when needed.
For a given LSP, give the order of alternative physical paths that support it. For example, a list of paths may be established. The first path in the list is considered the main path. If the main path fails to be established, the second path in the order list is attempted to be established.
Re-optimization of the LSP is allowed or disallowed when not working.
Define a set of resources for the physical path of an LSP that must be explicitly included or excluded. A resource group can be regarded as a "color" assigned to a link, and a group of links with the same color belong to the same class. For example, a network policy may specify that a given LSP cannot pass a golden link.
Establish LSPs in order of priority. This allows the LSR to first establish an LSP with a higher priority and then establish an LSP with a lower priority.
Determines whether an LSP can preempt another LSP from a given physical path based on the attributes and priority of the LSP. Preemption allows the network to revoke an existing LSP to support a newly created LSP.
By using constraint-based routing parameters, a solution to the LSP layout problem is automatically obtained.
Access billing and service statistics at the level of each LSP. These statistics can be used to characterize business, optimize performance, and plan capacity.

Juniper network traffic engineering structure can provide some advantages over the current IP-over-ATM model:

Supports high-speed optical interfaces.
The explicit path allows the network administrator to define the exact physical path of the LSP through the service provider network.
Supports dynamic failure recovery to a pre-calculated, hot-standby backup LSP.
Because LSPs are very similar to connection-based virtual circuits, LSPs can be used directly with existing offline network planning and analysis tools. The output of these tools can be translated into settings that establish LSP physical paths.
The statistics of each LSP will be used as a tool for future network expansion planning and analysis to analyze network bottlenecks and trunk utilization.
Constraint-based routing provides many enhanced features that allow LSPs to meet specific performance requirements before they are established.

In addition to supporting and extending the advantages of the coverage model, Juniper Networks' traffic engineering structure avoids the limitations of the existing coverage model's scalability issues and allows ISPs to expand their networks to OC-48 and above rates:

This structure provides an integrated solution that combines the Layer 2 and Layer 3 networks in the coverage model into a single network. This integration avoids the administrative burden of coordinating two separate networks, allows routing and traffic engineering to occur on the same platform, and reduces network operating costs. In addition, the LSP comes from the IP state, not the Layer 2 state, so the network can better reflect the needs of IP services. As ISPs continue to grow, Juniper Networks' traffic engineering strategy provides the same functionality on a single integrated network, eliminating the need to order, configure, manage, and debug two different sets of equipment.

This structure will not be limited to OC-12 links due to technical challenges in developing OC-48 rate ATMSAR router interfaces. This means that the lack of a high-speed ATMSAR router interface does not prevent ISPs from increasing their network speed to OC-48 or higher.
Because ATM is no longer needed as a Layer 2 technology, cell tax is completely avoided. This means that 15 to 25% of the bandwidth previously occupied by the ATM letterhead can now be used to carry other customer services.
The routing core network based on MPLS will not have the problem of ATM-like "N2" PVC fully closed network connection, so it will not cause pressure on IGP, which will lead to complicated setup problems. Modern Internet backbone network routers no longer have the performance problems that make ISPs configure fully closed networks in the first place to ensure network performance.
This structure does not require special Layer 2 technology (ATM or Frame Relay) to support switching and virtual circuits. Traffic engineering can therefore be provided at the third layer, supporting mixed media networks and reducing IP and "

While ISPs strive to keep up with growing Internet traffic, flow control has become a very important tool for ISPs. To enhance readers' understanding of traffic engineering and its important role in supporting the future Internet, this white paper will begin with how to implement traffic engineering in traditional router-based backbone networks. Then discuss today's

Over the past few years, the Internet core network has experienced exponential growth. Today, fast-growing traffic forces some ISPs to double their network capacity every three months. The successful ISPs who continue to increase their market share in a constantly changing environment are those who have the insight and flexibility to move their backbones to new technologies that meet the needs of growing customers.

In the early 1990s, ISPs relied on the use of metrics to manage the distribution of traffic flows through router core networks. Metric-based flow control provides a satisfactory traffic engineering solution. Until the mid-1990s, the backup capacity of the core topology began to limit the scalability of the solution. At the same time, around 1994 or 1995, ISPs needed to grow their networks, configure wider channels, and get deterministic performance from intermediate systems. At this time, the OC-3 and OC-12 interfaces of ATM core switches and routers appeared, and they can provide the required bandwidth. This became an important turning point in the development of the ISP market in the mid-1990s. ISPs who are aware of the limitations of the existing structure and moved to the coverage model by redesigning their networks can smoothly expand their market share and increase profit margins.

In the late 1990s, ISPs faced choices again when they planned to upgrade their networks to OC-48 or higher. Continuing the IP-over-ATM model, the result is considerable expense and increased complexity. Juniper Networks' traffic engineering structure provides business management performance of the ATM core network, while avoiding the limitations of ATM performance and scalability. ISPs who take into account the limitations of existing IP-over-ATM solutions and the advantages of MPLS / RSVP alternatives can understand that their past traffic engineering decisions will affect the future growth and profitability of their networks