Andrew Buss, originally published on The Register
With the worst of the recession looking to have passed (touch wood), things are looking up and growth is returning, even if it is erratic.
Companies are hiring again and people are buying more, resulting in an increasing pressure on systems and applications. The past couple of years have seen budgets slashed and renewals put on hold. As demands ramp up but budgets lag, should ageing hardware be put out to pasture or can it still perform a useful role?
Apart from the obvious implications of ageing hardware and relative performance, the issues of power consumption and space have an impact on server life extension. If either is an issue in your environment in terms of available supply or cost, the choice will lean towards decommissioning and recycling earlier rather than later wherever possible and spending the money on more efficient hardware.
If on the other hand you have little direct responsibility over the power budget, the power supply is ample and space is plentiful then the choice to extend the life of your older servers may be an attractive one.
Refurbishing servers to extend their working life is an obvious option in the quest to sweat server assets. But just how viable is refurbishment in reality? If refurbishing involves upgrading performance then the answer is most likely a resounding no. Upgrading single components is likely to result in a performance bottleneck, so CPU, memory, disk and possibly even networking may all need replacing. These items are usually expensive to obtain, and choice may be restricted in terms of socket type or interface. Factor in the labour involved to fit the kit, test it and possibly reinstall applications, and that upgrade no longer seems good value.
If upgrading is not cost effective, could utilising a pool of old servers work? An example could be taking a rack of 40 or so identical or generally similar 1U servers, and removing half from active duty, to be cannibalised for parts as they fail on the active pool. This would keep the active servers on duty at their existing performance level while extending their useful life without having to shell out for new parts or extended maintenance.
As to whether this option is viable, there are many factors to consider. Chief amongst these are: the cost of managing the inventory of available parts; the frequency of failures impacting both labour to fix and possible cost impact of downtime; whether the old servers are impacting necessary change elsewhere in the infrastructure; and if the servers are able to provide an acceptable level of performance or service. If viable, then refurbishment by cannibalisation may be an effective tool to gradually manage old servers out of the data centre.
If you are keeping old servers on board for that bit longer, there is a further choice to be made. Do you continue running the same configuration of OS and applications (in which case things can most likely stay as they are) or do you change the role of the server? Basic repurposing may be deploying a set of applications to the target servers that can run with sufficient performance to fulfil the required level of service. But manually deploying applications in this way can prove to be expensive and time consuming.
The best bet for repurposing may well lie in using the older servers as resources in a virtual server pool. In this way the old servers may be left powered down as much as possible as workloads are concentrated on newer, more power efficient and higher performance servers. If demands on these servers result in key applications experiencing marked deterioration in service levels then virtual machine management tools may fire up the resting servers and migrate lower priority workloads from the overloaded servers to the older servers. Of course, the ability to use these old servers as part of a virtual resource pool will lie in the hardware capabilities to support the required features. Older hardware may not support the direct migration of virtual machines from newer hardware and for servers that lack hardware support for virtualisation there may be both compatibility and performance issues.
An alternative use for old kit is to migrate it to an emergency disaster recovery site and have it available as a virtual server host for recovery purposes to provide rapid access to essential applications should the worst happen. This would allow access to a performance-degraded but available set of applications while a full recovery is implemented.
This dynamic workload management is a lofty goal, and may not be as easy to achieve in practice as many would like. But with application performance monitoring tools, network intelligence and load balancing technologies coupled with virtual server pool management, it is achievable. If there is an initiative ongoing in the data centre to implement advanced virtual machine and workload management on newer servers, extending it to the old estate where possible will help in repurposing old kit.
Once all other options have been considered and discarded, it is time to bid farewell to your old workhorses. Recycling regulations such as WEEE in the EU or the Resource Conservation and Recovery Act in the US mean that out of sight is not out of mind. Correct disposal is important for recycling and reuse and potentially some cash recovery. Hardware vendors have built up substantial recycling capabilities. Make use of these, particularly when they may also offer discounts or rebates for trading in old kit, or even offer buy-back of old kit.
Whatever the strategy regarding old servers, decommissioning hard disks and the data they hold should be handled in a formal manner. Servers should not just be removed from the rack and moved on, for recycling, resale or donation. The risk of data loss is too great for the potential cost benefit of recovering some cash by including the disks, even if they have been reformatted or repartitioned. Instead, the disks should be removed and either destroyed or subjected to secure erasure. And even once subjected to secure erasure, it may be advisable to retain the disks as internal spares rather than to dispose of them externally in a working state.