Tuesday, February 12, 2019

When Are On-Premises Systems Justified?

There is near-universal agreement that cloud computing is the future for enterprise IT. Our research at Computer Economics certainly indicates so. In just one year, our annual IT spending survey showed the percentage of IT organizations with 25% or fewer of their application systems in the cloud declined from 72% in 2017 to 61% in 2018. We expect a further decline this year.

Four Factors Favoring On-Premises

Even though the trend is strongly in the direction of cloud, are there situations where on-premises deployment is still justified? In a recent article, Joe McKendrick outlines four situations where staying on-premises may be preferable to cloud, at least for now. He writes:
To explore the issues of when staying on-premises versus cloud makes sense, I asked industry executives about any areas that were not suitable for cloud, and better left on-premises -- especially from the all-important data perspective. The security implications, as well as geographical presence requirements, are obvious. But there are also other facts that may make staying on-premises the most viable option.
Joe goes on to outline four factors:
  • Legacy entanglements: where the system is just one part of an integrated set of applications, especially where there are dependencies on certain database or platform versions. “Monolithic legacy applications” with custom system administration tools are another example.
  • Cloud sticker shock: where data storage requirements are so great that cloud deployment is simply not economical.
  • Security: where “some data cannot risk even a hint of exposure.”
  • Need for speed: where large data sets are maintained for “real-time user data interaction, high-speed analytics, personalization, or recommendation.” Some IoT applications may fall in this category. 

The Four Factors Not as Great as They Once Were

While these four factors are worth considering in a cloud vs. on-premises decision, I find them to be less of a factor than they were even a few years ago.
  1. The legacy system factor is certainly reasonable in some situations. To this I would add, staying on-premises may be justified when requirements for a new system can more easily be accommodated with an add-on to the legacy system. Be careful with this, however, as this can be a prescription for further entrenchment of the legacy system.
  2. In my view, cloud sticker shock is only a factor for a small percentage of cases, perhaps for very large data sets. Declining costs of cloud storage should lead to fewer instances where this is a legitimate objection. Often, IT leaders making a case for on-premises systems based on cost are not factoring in all costs, such as the cost of personnel to maintain and back up that on-premises storage.  
  3. The security factor I find to be largely an excuse. Although business leaders often underestimate the impact of a potential security breach, they also tend to overestimate the capabilities of their own security staff members, processes, and technology. The level of security maintained by internal IT organizations is usually far less than what is achieved by cloud services providers. If one of the big three credit data providers (Experian) could not protect consumer data maintained on-premises, what makes you think that your security capabilities are greater?
  4. The need for speed, in some cases, may be a legitimate reason for keeping some systems on-premises. However, most enterprise applications do not have this requirement. Even manufacturing execution systems—systems with low latency requirements—have been successfully deployed by cloud applications providers, such as Plex. In other cases, local buffering of data may be possible to accommodate any latency between the local system and the cloud provider. In such cases, it may be better to make investments in high-speed data communications, with redundancy, rather than continue to maintain such systems in local data centers. 
There is one more factor in favor of on-premises systems: Where there are regulatory requirements that the organization demonstrate control over the production environment. This includes FDA-regulated companies where a system is used to support regulated processes, such as quality control in medical device or pharmaceutical manufacturing. Although it may be possible to meet the requirement in a multi-tenant cloud environment, many regulatory affairs professionals are more comfortable not fighting that battle. In such cases, it may justify an on-premises deployment or at least a single-tenant hosted deployment where control of the production environment can be more readily assured.

Cloud First the Best Strategy

As discussed, there are situations where a true on-premises systems may be legitimately justified, although the case is getting weaker year by year. Nevertheless, for most new systems, business leaders should be adopting a “cloud-first” strategy, even if "cloud only" is not practical for now. If there is a cloud solution that will meet business requirements, that should be the preferred path forward. The advantages of cloud systems, especially in terms of alleviating the burden of system upgrades, are too great to ignore. On the other hand, if no true cloud system meets business requirements, or there are other limiting considerations, an on-premises solution may be a legitimate option. But even then, we would prefer to see a hosted solution, in order to achieve some of the benefits of getting application systems out of on-premises data centers.

1 comment:

clive boulton said...

Frank, a pattern I have noticed....

Smartphone applications are really hybrids. The client application is on the phone, a personal on-premises app. The data is saved to the cloud, either directly or synchronized with the cloud to deal with latency and reliability. The application is typically kept up to date by a marketplace push delivery.

I know that when we converted our on-premises micro ERP suite to this hybrid pattern, issues such as ability to automatically roll fresh on-prem updates to PC desktops across multi-sites sent our customer satisfaction through the roof. We also handled automatic database updates and strictly kept to one code-line. Same as smartphone customers could choose when to trigger upgrades. Another aspect any customized reports or any executables were automatically backed-up.

I now see this pattern evolving in Google Maps on Android, where AI detects a regular commute and automatically downloads a map the smartphone to speedup response. I expect top drawer enterprise applications vendors will move to a hybrid combination superseding both pure online and pure onprem. The robust product engineering required to do so, I expect will trigger further consolidation in EnSW apps (probably needs $4 billion run rate to amortize R&D).