A class by itself: Class-based queuing can provide simplified quality-of-service guarantees for high-bandwidth networks

Internet service providers are being challenged by their business customers to meet aggressive service level agreements. Increasingly, these companies are realizing that traffic management is essential to meeting these quality-of-service requirements.

Class-based queuing is a unique traffic management solution that allows ISPs to deliver bandwidth-controlled QOS in creatively packaged services, which offer new revenue opportunities while reducing operating costs.

By using class-based queuing for traffic management, ISPs can build services that deliver policy-based Internet access bandwidth to their enterprise customers. At the same time, they can build services that deliver bandwidth or rate commitments as part of their Web hosting or other server collocation services.

Traffic management using class-based queuing also can be an important tool for reducing cost while delivering consistent QOS from a service provider's regional point of presence (POP), as well as for providing scalable Internet access to multi-tenant real estate facilities.

Down on the farm Originally developed by the Network Research Group at Lawrence Berkeley National Laboratory, class-based queuing offers a flexible approach to sharing link bandwidth across a hierarchy of traffic types. Each class can represent an aggregation of traffic or an individual connection.

The hierarchy of classes allows many communities or departments to share link bandwidth, with each community further allocating its share among its own applications or traffic types. Traffic flows, or "classes," can be characterized by multiple attributes, including Internet protocol (IP) address, IP range of addresses, protocol or port identification. Each class is then assigned a committed bandwidth rate with a set of bandwidth attributes that tune performance (Figure 1).

Priority levels and borrowing privileges are two key attributes of class-based queuing.

Priority levels allow higher priority queues to be serviced first within the limits of their bandwidth allocation so that the delay for more sensitive real-time traffic classes is reduced.

Borrowing is an explicit class-based queuing mechanism for distribution of excess bandwidth. A class with borrowing privileges may initiate a borrowing request when it needs more bandwidth.

If another class is not using its full bandwidth, synchronization features indicate that bandwidth is available. Borrowed bandwidth is granted to higher-priority classes before lower-priority classes.

Class-based queuing also can be helpful in managing server farm bandwidth. ISPs are aggressively building server management facilities as enterprises accelerate the outsourcing not only of their Web sites, but in many cases their full complement of Internet servers. These servers typically are located at a service POP as part of a server farm locally managed and maintained by the service provider (Figure 2).

These server farms are growing rapidly, with hundreds of servers located in some service centers. However, the level of demand for access to the network is rarely well-defined or easily measured. It is disturbing that in most environments, any individual server can consistently consume excessive portions of available network bandwidth. This free-for-all server access to the network constrains the ability of an ISP to either offer or deliver any service level commitments.

By implementing a bandwidth solution using class-based queuing, an ISP now can design a managed Web site or server collocation service that includes bandwidth commitments. In this application, the class-based queuing device becomes the traffic control point between the server farm and the rest of the network, allowing the service provider to manage the amount of bandwidth available-and billed-to each collocated server or Web host.

The ISP now can establish SLAs that reflect the amount of network access bandwidth a server management customer requires. For example, Company A could contract for 64 kb/s of access to its mail server and 256 kb/s of bandwidth to its Web server. Company B could request 1 Mb/s of access to be shared across all its servers.

The class-based queuing traffic management software also offers a new service opportunity with its borrowing features. At the most basic level, borrowing gives users access to idle or unused bandwidth based on their needs and their borrowing privileges. The service provider can create a separate service level based on subscriber authorization to borrow bandwidth.

For example, if one server immediately needs more bandwidth than its committed rate, it can temporarily borrow from either idle bandwidth or from a pool of bandwidth set aside specifically to serve customers with borrowing privileges (Figure 3).

Where ISPs previously had to meet SLAs by over-building network infrastructure, they can now flexibly control the amount of bandwidth and server resources required to service their customers. For example, instead of provisioning 30 Mb/s of access capacity for peak-period demand, 10 Mb/s of capacity may meet the committed rate needs of those same customers.

Alternatively, this same 30 Mb/s of capacity could address the bandwidth demands of more customers, increasing the revenue potential and return on investment in the existing network infrastructure.

With the appropriate tools, the ISP can monitor the actual bandwidth usage of a customer's server to assess the need for additional bandwidth. If Company A's server is constantly attempting to borrow more than its committed rate, and it sometimes fails to get the bandwidth needed, the customer knows it is time to increase the amount of bandwidth committed to that server.

Think regionally Solutions based on class-based queuing are equally valuable in other applications such as regional POP bandwidth management. In a time when best effort service was acceptable, ISPs could oversubscribe network resources to achieve a reasonable balance between cost and performance. However, the business community has clearly established the requirement for SLAs.

Many service providers have accepted this challenge by over-designing their infrastructures to accommodate peak traffic levels. This is clearly an expensive, near-term alternative.

Moving forward, these ISPs need to better control traffic as it enters the network. Bandwidth management is acknowledged as a key strategy for connecting business sites to the Internet. It is also expected to become an integral component of regional POPs that aggregate subscriber traffic before it is directed to the Internet backbone.

By using class-based queuing for traffic management in the regional infrastructure, ISPs now can flexibly deliver rate commitments among access customers and among locally managed Web sites and servers. With rate control as part of the POP infrastructure, regional bandwidth facilities now can be designed to meet steady state requirements instead of peak load requirements (Figure 4).

For example, a current POP facility might be designed for a peak load of 30 Mb/s, with a backbone uplink circuit cost of $4000 a month. With guaranteed rates to customers, the service provider can possibly serve the same customers with 10 Mb/s of bandwidth at a circuit cost of $1000 a month. In this scenario, the service provider would save $3000 a month, or $36,000 a year, at a single POP.

Alternatively, rate-controlled access can offer a better return on the access link investment. More customers per link, each with a guaranteed rate of access, would offer the opportunity to generate more revenue over existing network facilities.

Adding traffic management into the regional POP also gives the service provider an opportunity to build the backbone based on steady state requirements rather than peak level demands. This saves money and offers the opportunity for more consistent service across the backbone.

Multi-tenant Internet access Delivering Internet access to the small business tenants of large multi-tenant facilities historically has been a costly, one-at-a-time undertaking. While it is generally recognized that these business tenants have the same need for affordable bandwidth and flexibility as larger, multicampus enterprises, the ability to serve this diverse market has been beyond the scope of what many ISPs could deliver.

The commercial real estate market, however, has discovered a currently underused asset. Broadband access facilities have been widely deployed across major metropolitan areas, easily reaching most buildings and office parks in those areas. More recent construction also includes the infrastructure wiring needed to deliver high-speed local area network services within a building or even across an office park.

With this "smart building" infrastructure now in place, ISPs can cost-effectively provision and guarantee affordable Internet access bandwidth to the individual businesses of multi-tenant facilities.

Smart buildings and even smart campus sites now position service providers to cost-effectively reach a large and rapidly growing smaller business community. In the multi-tenant Internet access application, a broadband circuit from the ISP feeds an access device supporting class-based queuing located at the building site. The Internet bandwidth is then partitioned to deliver a committed or guaranteed Internet access rate to each individual tenant (Figure 5).

Tenants benefit from ease of access to low-cost Internet bandwidth; building owners and operators can differentiate their more prestigious and more modern smart buildings. And ISPs now can cost effectively deliver full Internet access solutions to the previously inaccessible smaller business community.

This advanced traffic management approach allows ISPs to deliver rate-guaranteed Internet access. With class-based queuing, ISPs benefit from a fully interoperable IP class-of-service solution, providing network managers the high-performance traffic shaping and management controls needed to flexibly classify traffic and guarantee bandwidth to any number of building tenants.

For example, the service provider can allocate 64 kb/s of bandwidth to one tenant as easily as it delivers 3 Mb/s to another. Once provisioned, the ISP can adjust the amount of bandwidth easily and quickly to meet the business requirements of each tenant.

Most business tenants expect to scale to higher Internet access speeds over time. In most scenarios, tenants would start with a guaranteed rate of service. Once operational, some would find that they need additional bandwidth, but only periodically.

These customers could improve service quality by moving to the next service level, which would allow them to borrow from idle or separately allocated bandwidth. The service provider could then use collected statistics to track the amount of additional bandwidth tenants are actually using and advise them when it would be most cost-effective to increase to a higher guaranteed service rate.

The service provider also has a compelling cost savings opportunity. By leveraging the cost of a single broadband access circuit across multiple tenants, the cost of service delivery promises to be much lower. For example, provisioning discrete 64 kb/s or T-1 circuits to each tenant is both costly and time consuming, especially when compared with the cost of installing a single broadband circuit with a single box that allows software-controlled service provisioning across these same tenants.

Another advantage is in the ability to rapidly recover the cost of the local access circuit across the multiple tenants of the facility. In some areas, the cost of a broadband T-3 rate circuit is approaching the cost of four T-1 circuits.

In those areas, the allocated local access charge for a T-1 circuit could be significantly lower than the cost of a discrete T-1 local access circuit. As broadband access circuit costs decline, the multi-tenant customer could see lower access charges.

The service provider could also offer a customer monitoring or management option to more advanced building tenants. With this service, tenants could access their own configuration and management information to monitor usage and further partition their bandwidth in any way that suits their specific business needs.

Implementing class-based queuing managed bandwidth services enables ISPs to create new revenue opportunities and deliver business quality class-of-service applications. Internet access SLAs are critical as the enterprise begins to outsource server applications and other Internet and intranet applications.

Because customers continue to require more bandwidth as applications migrate to the Internet, ISPs that can offer traffic management and scalability will be able to differentiate services and increase market share while optimizing backbone bandwidth usage and containing operational costs.