It constantly surprises me how often I see simply adding more hardware as the application performance strategy for tier 1 mission critical applications. The mantra normally is that hardware is cheap, and engineers are not, so adding hardware, even if somewhat of a band-aid solution could be a good strategy.
The challenge is that hardware will only take you so far – It is true that adding hardware will likely mask or delay fixing performance problems for a time, but is it really saving any money? And what of the risks?
The risk is of course that by simply throwing hardware at the problem (which could be your own hardware, e.g. web servers etc, or 3rd party like CDNs and so on) all you do is ignore the symptoms of any underlying bottlenecks in the application. While it is true that upgraded hardware is sometimes the best course of action, this should be considered as part of an overall capacity management plan, and as the outcome of a proper analysis rather than the first course of action.
What will likely happen is that that hardware will be underutilized because it does not address the fundamental bottlenecks that exist in the application (e.g. in the code, how it is deployed etc). The new hardware is masking these bottlenecks, so they are then likely to be apparent only during peak loads – the most important times for the application. It is these times that are likely to have the greatest business impact, which for many firms could run into the millions.
At this point in time the hardware is not looking so cost effective, especially as the organization who has simply thrown hardware at the problem is likely to be ill equipped to deal with the problem quickly as they will not have the visibility needed to isolate and actually fix the issue.
The hardware is also likely more costly than originally thought. Is the true cost of adding that hardware actually being measured? These costs include the purchase price, maintenance contracts, power and cooling, data center space, administration time, software licenses and so on as well as more intangible costs such as opportunity lost, VM Sprawl, environmental aspects and even companies who have to build new data centers because their current center(s) are now full.
Instead of hardware, what is needed is an Application Performance Management strategy. The strategy is not simply buying a tool, rather it is an overall approach to application performance which includes understanding the mission critical applications and their needs, measuring those applications appropriately, understanding business impact of poor performance, being proactive in terms of managing acceptable performance over time, as well as being ready to find and fix issues rapidly should they arise.
Over the long run, is hardware really the answer?
About Bryce Dunn
[...] [...]