/vnd/media/agency_attachments/bGjnvN2ncYDdhj74yP9p.png)
The phenomenal success of the Internet has motivated network and service providers to embrace IP as their routing technology of choice
Let us travel back to the not so distant past before the Internet became a major vehicle for communication. At that time, the telephone was still the major means of rapid communication. A customer could pick up the telephone, dial a number and talk to another. That connection stayed in place until it was disconnected and it was available 24x7. Customers came to expect this level of service which came to be termed carrier-class.
As we approach 2004, communications are quite different from the times of the telephone and the circuit-switched network.
The Internet has changed the way we live and conduct business. The Internet protocol (IP) has become the prevailing network protocol. Carrier requirements have increased significantly. While the requirements that existed in that pre-Internet world are still necessary, they are no longer sufficient. The specifics that defined carrier-class even as late as the late 1990s will not suffice today and what defines it today will not suffice in the latter years of this decade. In many ways carrier-class, as it must be defined to suit an IP network, is a moving target.
IP Networks
First, let’s look at the characteristics of an IP network. Internet growth during the past few years has been driven by several factors. E-mail has become an essential means of communication. The World Wide Web (www) has been accepted well by both business users and the general public. The use of corporate intranets and extranets has greatly increased. Internet usage has literally shaken the foundations of networking in both the data communications and telecommunications industries. Both industries are converging toward a common infrastructure that is based on IP protocol and packet switching–an IP convergence model.
The Internet of today is a makeshift infrastructure that has been built as an overlay above existing telecommunications and wide-area data communications networks. The amazing reality is that within a relatively short period of time, the network based on this ad-hoc infrastructure has become ubiquitous–accessible from virtually anywhere on the earth and by virtually anyone on the earth. Its crucial failing, however, is that it does not meet the reliability and performance standards of traditional wide-area data communications services. Today’s Internet infrastructure is severely restricted by its inherent inability to differentiate traffic and to deliver differentiated quality of service (QoS) to traffic from different categories of users and applications. While some applications such as e-mail might tolerate highly variable delays, more business-critical and time-sensitive applications demand stricter service provisioning. Providers offering such services (services requiring true QoS) need infrastructure solutions that can classify traffic, provision service, and maintain usage statistics for billing and planning purposes. The infrastructure must not only support these requirements, but also must do so without any performance degradation.
Requirements
Taking into consideration the qualifications to be carrier-class and the unique nature of IP networks, the following factors are necessary for a carrier-class IP network.
n Availability/reliability
n Redundancy
n Provision of IP services
n QoS
Reliability
That carriers require networks that are reliable, resilient, and dependable-networks is indisputable. The expectation is that to be carrier-class, an IP network must deliver the same level of availability and reliability that the existing circuit-switched network offers. Network components must be engineered with high availability, reliability, low failure rates, and rapid recovery time as primary design criteria. Ideally, any failure that does occur should be transparent to customers.
Availability and reliability refer to fault-tolerant hardware and software. Hardware should be redundant and hot swappable.
The reality, however, is that most network crashes are due to software failures. They are typically related to software upgrades, fixing bugs, installation of new releases, the addition of enhanced features etc. To minimize such failures, software must be modular such that the failure of any one software component will not impact the remainder of the system. Thus, downtime can be minimized even if a failure did occur. Networks of the past have principally relied upon big monolithic software architectures. By their very nature, these massive suites of software were difficult and time-consuming to test.
Attempts to shorten the testing process resulted in failures that showed up in the production environment. It should be noted that system monitoring and auto diagnostics of both hardware and software are on the ‘must-have’ list if carrier-class levels of availability and reliability are to be realized.
Redundancy
The redundancy (of physical components) eliminates single points of failure. Redundancy must be viewed in terms of hardware, software, the system–and the entire network. Within a network, redundancy should include both network components (such as routers and the components that comprise the routers) and multiple and diversely routed links and paths.
Software-based recovery mechanisms can complement redundancy by minimizing the impact of failures. For example, this can be done by automating the recovery process to ensure the fastest possible recovery, thereby making any failure transparent to the end user. Redundant hardware can increase network stability and performance by preventing reroutes caused by hardware failures.
Provision of IP Services
The phenomenal success of the Internet has motivated network and service providers to embrace IP as their routing technology of choice. That success, however, has resulted in requirements for increased bandwidth and in demands by customers for new applications that have very specific bandwidth, quality, and reliability criteria. But IP has not been designed with QoS as its paradigm. As a result, IP’s very success could well become the force that actually ends up degrading its performance.
Service providers face intense competition just to continue in business. The implosion of the dotcom and telecom industries since 2001 has exacerbated the problem. This very basic need to survive has forced them to give performance guarantees or SLAs to customers to differentiate their services from that of their competitors. IP services like IP VPNs can provide the competitive differentiator for these providers. Such services can provide the key to meeting customer demands, keeping up with competition, and generating new revenue. But with the provision of IP services comes the issues related to the inability of IP networks to implement QoS. The concerns related to prioritized and differentiated services become crucial. Carrier-class, therefore, must imply the ability to provide IP services, and it must imply QoS. However, as stated, QoS is something that IP does not do well.
Quality of Service
QoS is the process of treating and measuring traffic in terms of that traffic’s importance. Defining importance is simple. It’s an economic decision that is typically based solely on the willingness of customers to pay a premium for specific services and service levels. QoS refers to the capability of a network to provide better (or prioritized) service to selected network traffic.
There are ranges of different applications that operate across the network. Some have very strict performance requirements, and some have virtually no such requirements. To ensure network efficiency, every application used on that network must be considered in terms of its individual requirements to operate efficiently. For example, voice traffic has low delay and low delay variation; some loss is tolerable. Data traffic can tolerate a higher delay and higher variation in delay and can tolerate varying bandwidths. With video traffic, a higher delay may be more acceptable than it is with voice traffic. But data loss is unacceptable. All these criteria must be considered within the bounds of what is important enough to persuade a customer to pay a premium. QoS is really about providing a premium ‘business-class’ IP service over the Internet. Many service providers solve their QoS problem by simply throwing more bandwidth at the problem. While there are situations where this is an appropriate solution, it cannot be the long-term solution. If supporting SLAs is to be an objective of QoS, then performance must be predictable. Therefore, an objective of QoS is to provide predictable service to support SLAs. Metrics related to QoS include availability, latency, delay variation, throughput, and packet loss.
Summarizing how this relates to provision of services, we can conclude that QoS is really about the following:
n Minimizing delay
n Minimizing variations in delay
n Provision of consistent throughput (making IP predictable)
To sum up, carrier-class IP networking is ultimately a lot about QoS.
Alice Barrett Mack, president ABM Technical Communications Courtesy: IEC
(www.iec.org)