INTERNET TRAFFIC MANAGEMENT: A NEW GLOBAL PERSPECTIVE

Introduction

The clear advantages of the Internet and its attendant technologies have led to the proliferation of corporate Internet, intranet and extranet applications. Whether the application is a corporate presence, commercial application or internal resource center, more and more businesses are deploying network solutions that are being used by more and more people. With the phenomenal growth of IP traffic due to market growth or increased company reliance, server sites are quickly becoming overwhelmed with traffic. There are essentially two ways for server sites to manage increased traffic: deploy a more powerful server, or add additional servers and create a server farm. Both of these solutions present challenges.

The Roots of Distributed IP Traffic Management (Local Load Balancing)

The strong server solution is expensive, not scaleable, and requires service interruption for maintenance and upgrading. The cost of replacing a server with a stronger one is twofold. First, the existing server must be deactivated. Though perfectly functional, the server is rendered valueless to the site. Second, the new server must be able to handle all existing traffic and projected future traffic, and as such usually requires large capital outlays. Notice that this solution is not scaleable, that is legacy equipment must simply be discarded – every time upgrading is necessary. Notice also, that replacing the service necessitates taking it off-line for a period of time while the new server is installed.

The server farm solution is more affordable in that a new server is added to the existing server which retains its value. Also, the new server does not need to be powerful enough to handle all traffic alone, and as such is less expensive than the server required in the strong server scenario. Each time the site needs to be upgraded, a new server is simply added, without disrupting service. The challenge presented by the server farm solution is that each server has a unique IP address. This means that users must remember, or have access to, more than one address. It also means that the traffic is not distributed between the servers in an efficient manner. Load balancing devices that manage local traffic, like RADWARE’s WSD family of products, provide the solution to this challenge.

When an load balancing device is installed at a server site, it is assigned a Virtual IP address (VIP). Users access this IP address, and the load balancer routes traffic to the servers in the server farm. Additionally, the load balancer monitors the load on each server, and routes the traffic accordingly. This means that users reference the site through a single URL (IP address) regardless of the number of physical servers that actually exist. In addition, since an intelligent load balancer is employed they access the site in the most efficient manner. Furthermore, unlike the strong server solution, server farms do not have a single point of failure, so if a server goes down, there is another server available for the traffic. The load balancing device ensures that traffic is not routed to a server that is unhealthy, down or simply saturated with requests.

Deploying a load balancing device at sites consisting of multiple servers in a server farm enables access to the site’s application(s) via one IP address, masking the underlying site complexity. It also maximizes the efficient use of the servers, and protects against site inaccessibility by routing traffic away from non-optimal or downed servers. However, the meteoric rise of Internet business, and the reliance of large organizations on sophisticated intranets and extranets to provide services to employees, vendors and customers, has led to new challenges for server sites. Corporate and customer data must be available 24 hours a day, 365 days of the year. In order to meet this requirement, network administrators are being forced to locate their content in more than one physical site. Two issues necessitate this configuration. One is the need for a disaster recovery solution, in case a single corporate site becomes unavailable. The second is to send users to the best site in terms of both performance and network proximity. To meet these needs more and more corporations are deploying multiple data centers (server sites) in disparate physical locations across cities, states and countries. Local load balancing alone cannot handle this topological challenge. It has become imperative to balance IP traffic amongst servers, and server farms, located great distances from one another.

Global IP Traffic Management:

There are a number of reasons why an organization might find itself with server sites in multiple locations. The most obvious scenario is the Web company whose service is provided solely via the Web. In this case, the company might co-locate its Web site to ensure that the site is close to its users who may access the site from anywhere in the world. Or the company may co-host to ensure that the site remains impervious to natural and man made disasters, such as ice storms or sloppy road crews, who may cut the very wires bringing IP traffic in and out of the region.

However, it is not just Web centric organizations that have server sites distributed among numerous physical locations. Small and large organizations, from 30 to 300,000 employees, realize increased productivity through resource sharing among different branches located in various locations. Though each office may act as an independent business unit, resource sharing among company branches leads to better and smarter buying, selling and administration. Additionally, large organizations might have to provide all employees, regardless of the city or country in which they work, with solutions and critical resources that must be available in a timely fashion, worldwide. To ensure each business unit’s independence and competitiveness, organizations are deploying complicated intranets and data centers to which contributions are made real time worldwide, and conversely data is extracted 24 hours a day, seven days a week.

The globalization of business practices has led to the need for global server sites, and appliances that can ensure availability and efficiency over great distances. Whether a company must connect disparate offices, provide services to customers around the world, or host global events such as elections or sporting events, quality global load balancing appliances ensure availability, efficiency and speed.

What To Look For in Global IP Traffic Management Solution:

Like their "basic" load balancing forebears, global load balancing appliances must provide a range of services. Collectively, these services provide site optimization and availability, but do so over geographic distances in a way that is entirely transparent to the user, irrespective of the nature of the use. Furthermore, a global load balancing appliance must be flexible enough to deploy a broad range of Internet traffic management methods in order to effectively handle the organization’s evolving business needs. As the organization’s traffic management needs change, its method of load balancing may also need to change. A good global load balancing solution offers many methods in one box, thereby protecting investment and reducing network administration. Global load balancing (GLB) appliances must:

• Treat corporate resources as a single entity. Functionally this equates to being able to represent diverse physical data entities via a single URL. Users may still want multiple entry points into their networks with multiple URLs. They may, however want the entire corporate data resource to be represented by a single logical address that physically corresponds to entities at multiple physical locations. This requires the GLB to be flexible enough to provide single or multiple logical entry point(s) into the network. For example, Acme.com may have server farms in 10 different locations but want to logically represent all 10 locations, and intelligently load balance between them, through the single URL www.acme.com. Physically there may be 200 servers distributed among the 10 farms, but the end user would see this entire entity through the single URL. Once again, but on a much larger scale, the load balancer is hiding the underlying network complexity from the user by representing a large complex corporate network through a single logical entry point.

• Provide for distributed site failover and redundancy. Many users want to use global load balancing to provide total site redundancy. They want load balancers and device farms at multiple sites and also want those sites to back each other up. They expect the global load balancer to immediately detect the failure of one site and divert user traffic to an active site. In this context the load balancing solution must work seamlessly within the existing DNS structure of the Internet. In terms of configuration flexibility, the user may want all sites to be active or he may want to designate one site for disaster recovery, making it hot only if the primary site goes away.

• Be able to load balance globally based on real time load and/or network proximity. Load balancing based upon real time load, between different physical sites, requires all the load balancers in the network to continuously update each other with health and performance indices. Each site must know the condition of every other site if it is to make the appropriate redirection decisions, based on real time dynamic load. Network proximity is also not the same thing as geographic proximity. Network proximity is a measure of how far away a user is from a data resource in terms of network distance or time, versus physical or geographic distance or time. For example, even if a user is geographically closer to New York City than Chicago, he may be able to get to Chicago much quicker if the network path to New York is slow or overloaded. In terms of network proximity, the user is closer to Chicago than New York. A GLB solution must continually monitor the real time status of the network and respond to permanent or transient performance conditions, directing customers to the "best" site.

• Provide measurement and redirection flexibility. When looking at global solutions, it is important to ensure that the GLB has flexible and configurable decision making capabilities. Redirection solutions based upon DNS alone may not result in a client going to the "best" site for that client. When selecting a GLB solution, look for one that offers multiple redirection mechanisms as well as multiple methods of measuring network proximity and latency.

• Be able to provide both local and global load balancing. All of the load balancers within the network should be capable of communicating with one another to share dynamic load as well as site and server failure conditions. This feature, like many others, is very vendor specific. Some vendors sell one device for local load balancing and another device, often sitting in the same rack, to accomplish global redirection. These limitations will further complicate creation of a redundancy scheme between the load balancers themselves. A strong load balancing solution will be able to provide both local and global load balancing capabilities within a single device. A GLB should offer all of the advantages of a local load balancer, while extending this functionality to the entire global business unit.

• Be able to scale with the corporate organization. That is, the load balancer used today to accomplish local load balancing should be able to be simply and easily upgraded tomorrow to accomplish global load balancing. Also, the GLB must demonstrate flexibility by offering an array of traffic management schemes. By doing this, the GLB is not limited to performing traffic management according to today’s requirements. If, for example, an organization that currently redirects traffic in the DNS resolution stage needs to redirect HTTP traffic as well as, or instead of, a high-quality GLB will allow it to do so seamlessly and transparently. A superior GLB is able to adapt to the specific and changing needs of growing and dynamic business units.

Conclusion

A global load balancing solution is about increasing scalability, optimization and fault tolerance. It is about ensuring high availability. With a global solution, a company’s assets can be represented logically by a single URL even though physically those assets may be distributed throughout the world. A strong global traffic management solution will also provide for disaster recovery and geographic failover between sites.

This becomes extremely important if the content you are publishing needs to be accessible 7x24 (remember the ice storm and road crew mentioned above). A distributed solution should be able to recognize this type of scenario immediately and divert users to an operational facility. It must also ensure that local traffic is balanced among servers in a local farm, optimizing each farm’s performance. By handling local and global Internet traffic, and by doing so according to a business unit’s unique needs no matter how they might change over time, a good global load balancing solution ensures present and future optimization of asset availability in a cost effective and confident manner.

RADWARE introduced its first load balancing solution, the Web Server Director in Q4 of 1996. Since then it has consistently won industry acclaim as a leader in load balancing technology. Today RADWARE (Nasdaq: RDWR) is the only company to provide a complete IP traffic management solution for servers, firewalls, cache servers and routers. RADWARE routes all IP traffic. For more information on local or global load balancing visit www.radware.com.