Original URL: http://www.theregister.co.uk/2011/04/19/rackspace_cloud_load_balancers/

Rackspace fluffs up load balancing for clouds

Goodbye HAProxy and Nginx, hello virtual Zeus

By Timothy Prickett Morgan

Posted in Cloud, 19th April 2011 19:52 GMT

The whole point of cloud computing is to make infrastructure not only easier, but as transparent as possible. And so Rackspace Hosting, which wants to dominate commercial cloud computing, has added a much-needed feature to its cloudy infrastructure: integrated load balancing.

Load balancing is nothing new among Rackspace's managed-hosting customers or those using virtualized slices on the Rackspace Cloud. Josh Odom, product line leader for the Rackspace Cloud platform, tells El Reg that companies use a number of open source load balancers, such as the open source HAProxy and Nginx load balancers that are popular among its hosting and cloud customers.

These load balancers distribute workloads across multiple machines when a customer's applications outgrow one server (or one virtual server slice), and are also used to provide high availability for applications so that a server crash doesn't knock them offline. And because a load balancer is a single point of failure, customers almost always have not one, but two.

If there is anything more annoying than setting up a load balancer, it is having to set up two of them. "It is a very complicated configuration for a system administrator," says Odom. "Many people set a load balancer up and they don't even know it isn't working properly until something fails."

After reviewing a slew of load balancers – including hardware appliances and software appliances – Odom says that Rackspace chose Zeus Technology's Traffic Manager to be the virtual load balancer for its cloudy infrastructure and managed hosting customers alike.

Rackspace plunks Traffic Manager on a bunch of Linux boxes in its data centers, and exposes it through the configuration screens for its cloud and hosting setup. Customers select the cloud or managed servers that they want to put on what IP addresses and get their workloads balanced "in 30 seconds," as Odom put it.

The load-balancing service can be used to split workloads across company sites or between Rackspace and other public clouds, too. Any public IP address can be added to the load balancer and have workloads distributed to it.

Several hundred customers have been beta testing the load-balancing service for the past several months, and a number of them were using it to migrate workloads off other service providers' iron and onto Rackspace machinery in a controlled fashion. (Of course, the load balancer could be used just as easily to move workloads off Rackspace iron, too.)

Rackspace charges 18 cents per gigabyte per hour for outbound network traffic and 8 cents per gigabyte inbound for network services. The Cloud Load Balancer service costs an additional fee on top of that. Rackspace watches the number of concurrent connections per hour hitting the load balancer, and averages them for each hour. It costs 1.5 cents per hour for every 100 concurrent connections on top of the regular bandwidth fees.

The way it works out, for most Rackspace Cloud customers, it is something on the order of $10 to $15 per month in incremental cost. And now the customers can nuke two of the physical or virtual servers, which were running other load-balancing software – and they no longer have to configure and maintain it.

As of this morning at 11am eastern, only two hours after the product was announced, 700 customers at Rackspace had already put Cloud Load Balancers into production. ®