Get out of the habit of buying a server for each new application and consolidate.
It is no surprise that business departments like having their own applications, running on their own machines, bought out of their own budgets and under their own control.
Wintel/Unix machines are cheap and one application per server is simple as there is no risk of applications interfering with each other. But over the years servers have bred like rabbits and now uncontrolled server proliferation has led to costly and unmanageable chaos. Ironically perhaps, to impose order on this chaos, the Wintel/Unix world is turning to technologies and management philosophies that were developed decades ago in the mainframe world.
When servers are counted in their hundreds, datacentre space becomes an issue, as does providing disaster recovery. Moreover, with average utilisation rates running at a mere 10%, the total unused capacity of a large server estate is a massive waste.
That level of waste has never been tolerated in the expensive mainframe world. Running multiple applications on one machine has always been the norm, so facilities such as OS390s VM environment and logical partitioning were developed. Now, similar techniques are under development for the Windows world. Microsoft will enhance Windows to automate the deployment of systems and provide the ability to dynamically allocate server, storage and network resources based on business demands.
As well as matching the raw computing power of mainframes - eight, l6 and 32-way servers are commonplace - virtualisation means that multiple applications can run on one server. However, each application will need its own copy of the operating system and therefore its own licence, There can also be problems with supplier support for software running in a virtual server.
Additionally, virtualised IO can be slow, as each virtual machine can only take advantage of a single processor. Clustering is only supported within a single host server, not between remote servers. For these reasons, virtualisation works best for development purposes, for isolating badly behaved applications that might interfere with other applications and for running applications that use older, unsupported operating systems.
Complementary to virtualisation is mainframe-style co-hosting, where multiple applications run under a single instance of the operating system. This works well for consolidating large enterprise resource planning packages and databases as well as many smaller applications. However, large applications must be capable of scaling up on SMP (symmetric multiprocessing technology.
But scale-out applications such as web servers may consolidate better on blade servers, as will applications where a dedicated server is required - for security, for example. However, blades should be used sparingly or the cost of management and administration will spiral upwards.
As well as new technology, consolidation requires changes to datacentre management pro-cesses - test and release must include consolidation testing to check the application can co-exist with already installed applications. Capacity management, workload management and tight configuration control is vital. Methods of user chargeback may need to change and there may be cultural resistance to centralised consolidation to overcome.
When embarking on a consolidation project you cannot just virtualise the problems away. First, build an inventory of existing servers and applications and analyse them to identify groups of applications that can co-exist without interfering with each other - ideally using automated tools.
Next, develop a process for profiling and prioritising applications. Start with relatively easy targets such as databases and then move on to more difficult "legacy" applications in parallel and focus on consolidation.
Because each consolidation technology has its strengths and weaknesses, a deployment strategy is required. Develop a technical architecture for consolidation using a mix of technologies and draw up architectural rules for selecting the most suitable deployment platform for each category of application.
Get out of the habit of buying a new server for each new application and instead purchase servers as the requirement for capacity grows. Finally, as the numbers become more manageable, it becomes feasible to apply true enterprise-class datacentre management standards to your consolidated server estate. If you do nothing, those servers will keep breeding.
Jim Wright is a systems architect at Unisys Technology Consulting Services