Load Balancing on a Shoestring

If you own or manage a high traffic or mission critical website then some sort of load balancing solution should be implemented. A few ways to implement a load-balancing network are: employing layer 4 routers, DNS Round-Robin, and a Linux Virtual Server. Each of these should be implemented for different situations and needs.

 

A layer 4 router operates at the Transport Layer of the OSI model. This means that it operates above the IP layer which is contained in layer 3. Basically what happens with a layer 4 router is that it disguises multiple servers behind it to appear as if they were simply one server. This effectively creates a high-availability cluster that allows for one server to go down and be removed from the rotation without anyone from the outside knowing.

There are several vendors of Layer 4 routers, however almost all of them are extremely costly. Foundry Networks , F5 Networks, Cisco, and Coyote Point all offer layer 4 routing solutions. To implement one of these solutions there are several things that are needed to assure high availability. A layer 4 router, multiple front end web servers, and clustered or mirrored database servers are all needed to provide an effective load balancing solution with a layer 4 router. The downfall of this solution is its prohibitive cost. These solutions are usually only obtainable by large companies with big IT budgets.

A more affordable alternative to implementing a Layer 4 router is to create a DNS Round-Robin. This type of load balancing is done at the DNS server level as opposed to the router level. What happens is multiple IP addresses of servers are entered into the DNS server. Then when the first person loads a page they will get the first IP, the second person will get the second IP, and so on. When the last server is reached, then the DNS server starts again at the front of the list.

To implement a DNS Round-Robin load balancing solution, you need to create multiple identical servers that have unique IP addresses. Then list all the IPs in the zone file of the DNS server. Now the servers will be called in a distributed fashion. This setup is much cheaper than the Layer 4 router approach as it does not require a router that can cost thousands of dollars.

Unfortunately the phrase “you get what you pay for” holds true on this setup. A DNS round robin approach has many drawbacks. The first drawback is that if any server goes down, a portion of the traffic will still be sent to it and they will see a broken server. If a server permanently goes out of commission it will take a while for it to be completely removed from the loop due to DNS propagation times. It is also not true load balancing since there is no way for DNS round robin to control which server gets the next connection based on knowledge of existing number of connections and load. The best application for a DNS round robin is a high traffic static or semi-static website.

Despite the problems with DNS Round-Robin, there is still a pretty good solution for creating a load-balanced system on a budget. This dream solution that merges both cost and high availability is called the Linux Virtual Server. The Linux Virtual Server is a solution that can turn a server using commodity hardware into a layer 4 router. One of the advantages of this over using a layer 4 router from a major vendor is that if any part dies it is relatively easy to replace in minimal amounts of time. For instance, spare network cards, RAN etc. are very common in datacenters but a spare proprietary interface card is pretty rare.

 

To setup a Linux Virtual Server ‘cluster’ you will need similar equipment that was required for a layer 4 router. If you want a high availability setup, a distributed database server is important. However, if the point of load balancing in your organization is to simply add more bandwidth capacity to your servers, then one database server may work fine. More information on setting up a Linux Virtual Server can be found here.

While the Linux Virtual Server may sound like an ideal solution, it too has a few drawbacks. The biggest drawback is that it is a Linux system acting as network hardware. Running a Linux machine at that level requires some expertise, and you will need a knowledgeable person on staff to manage this router. Luckily, most organizations with the kind of website that requires load balancing do have a Linux guru or two on the payroll.

There are several ways to create an effective load balanced network for your website. Most of these have a few drawbacks such as price, lack of robustness, and maintenance. If you are in charge of maintaining a high traffic or mission critical website then one of these three load balancing methods will probably work for your site.
 

by Tyler Weaver