Compute Pools Feature Story

Azul Systems solves the biggest challenge facing IT manager today - to reduce the cost and complexity associated with delivering compute and managing enterprise class data centers.

The Challenge

Today, data center compute power is typically purchased in a “denomination.” 2-way, 4-way, 8-way, etc. servers are selected depending on the demands of a specific application. Every time a new application is deployed, IT managers must decide how much compute power is necessary to run the application and meet required service-level agreements. In order to ensure that each application can meet peak demand, the sizing decision includes significant over-provisioning.

Capacity planning guidelines typically assume a 5-to-1 peak-to-average ratio, then recommend adding at least a 100 percent safety margin for errors in estimating future peaks. This means that if the average utilization goes over 10 percent on any application’s available resources, a likelihood exists of failing to meet capacity needs during peak usage. If the average utilization reaches 20 percent, failure during peak usage would be common.

Many data centers running mission-critical applications play it safe and run application servers at a sustained average utilization of about 10 percent. This means millions of dollars are spent to deploy and maintain servers with low utilization resulting in slow return on acquired infrastructure and very high total cost of ownership.

As data centers have evolved over the decades, disparate and completely incompatible platforms have been installed, making data center-wide server consolidation extremely challenging.

The time has come for a new “non-denominational” approach to delivering compute resources—one that eliminates capacity planning at the individual application level while also addressing this ongoing problem of “heterogeneity.” Enter the era of network attached processing.

The Azul Solution: Network Attached Processing

Network attached processing is designed to “unbound” compute resources for applications running on Java™, J2EE™ and other virtual machine-based platforms. The idea is to “separate compute from the computer,” a concept that has strong roots in technology evolution. Back in the 80s, servers did just about everything. One of their growing functions was routing network traffic. In the mid-80s, Cisco popularized the router and leveraged the TCP/IP open standard, a device specifically designed to manage this function for the server. This allowed devices to share the network more easily and larger networks to be built. Now nearly all network traffic goes through routers and other optimized networking equipment. In the 90s, network attached storage (NAS) was the result of a similar evolution—storage capacity was moved from the server to the network. Now, Azul Systems has taken the evolution one step further by moving compute resources from the server to the network, where it can be made available to any VM-based application that needs it. Read more »