Is The Cloud Slower For Analytical Insights?

In my last post I described a big problem related to data in the pandemic. According to an IBM global survey of information executives, the Covid pandemic, like previous business crises, put pressure on IT organizations to deliver business information quickly. Reliable real-time data and analytics for decision-making became more critical. But many IT organizations couldn’t deliver; survey respondents said that the information and insights needed by the business often weren’t available. They mentioned issues of timeliness, integration, and a shortage of IT services as reasons. The IT executives agreed that real-time analytics for decision-making are particularly critical during business crises, but many can’t provide them in real time. Instead, their systems and reports are more focused on transaction data, which tends to be provided more quickly than analytics.

In this post, I’ll try to address some of the underlying causes of poor data and analytics availability during business crises like the pandemic—because we’re likely to see more crises in the future. And even without a crisis, as companies become more dependent on data-driven decisions, business leaders are going to want reliable, rapidly-delivered insights from analytics and AI.

Is the Cloud Slower for Analytics?

One factor in the failure to deliver real-time analytical insights is the shift to cloud architectures. If this isn’t controversial enough, I should probably deliver another trigger warning: I will argue that there are some circumstances in which on-premise solutions are better for this purpose.

This idea is heretical because moving data and applications to the cloud is taking on aspects of a popular religion. No doubt there are many good reasons and use cases for cloud migration, economic and otherwise. But I try to remain agnostic on the issue of premise vs. cloud, and for that matter whether my computing cycles are delivered by a mainframe, a server, a laptop, or a Raspberry Pi. I prefer to work with vendors that offer multiple options as well. The survey I am quoting was sponsored by IBM, which has a strong premise-based business as well as a growing cloud business, particularly in private clouds.

By the way, I feel the same way about electricity. Most of the time I just plug into the electrical cloud, but the recent unreliability of the cloud electricity in my area during bad weather just prompted me to add some premise-based electricity—in other words, I installed a generator. Of course, I still pay my electric bill and generally rely on the electrons that flow from my electric utility company. Similarly, if I ran my business on the cloud, I’d probably continue to use some cloud services, but some recent cloud outages might provoke me to install some on-premise compute and storage.

Preferring the Cloud, Even When It Doesn’t Make Sense

Not surprisingly given the current ethos, when the IT executives in the IBM survey were questioned about cloud vs. premise-based in terms of various attributes, there was some variability, but many preferred the cloud. However, they did evince some dissatisfaction related to speed: 67% agreed that “My organization finds that storing or processing some of our information in the cloud can delay its availability for decisions that need to be made rapidly.” It is surprising that IT executives are willing to put up with such a delay, since 86% agreed that “data used for analytic insights is more time critical to my organization than other data.”

One survey question asked an additional question of the 95% respondents who agreed that a business or societal crisis highlights the need for a flexible and responsive reporting and analytics environment. It was, “Which do you feel is the best way to deliver this?” The results were that 59% said a combination of cloud and on-premise, 29% chose the cloud, and 12% chose on-premise. I certainly understand why respondents chose the combination approach, but I wonder why—if 2/3 of respondents believe that the cloud slows down analytics—29% of respondents prefer the cloud for analytics. In fact, of those respondents who believed the best approach is cloud based, 69% agreed that the cloud can delay availability of data for decisions. Clearly the rapid, responsive delivery of information is not their highest priority.

The survey results bear this out. When asked which of several attributes were most important for analytics, latency was the lowest:

• Security 78%
• Performance/cost 68%
• Resiliency 61%
• Latency 49%

There were high levels (95%) of agreement with the very rational statement that, “My organization makes decisions on where/how to process data for analytics and AI based upon the best platform attributes (latency, security, resiliency, and performance/ cost).” However, the outcomes for analytics latency suggest that the most rational architecture decisions are not being followed—only 14% preferred premise-based solutions, even though they often yield faster analytical insights.

A couple more bits of seeming irrationality: 74% agreed that, “There are downside privacy and security risks associated with deploying sensitive personal data in the cloud versus keeping it securely on-premises.” Further confusing the issue, 84% agree that “My organization is in a highly regulated industry that requires key customer data be processed on-premises.”

Yet most of these IT executives still prefer the cloud over premise-based analytics. Perhaps they believe that the cloud has enormous cost advantages over on-premise systems, but my impression is that this is only true under certain circumstances, such as a short-term need for lots of compute or storage.

Separating Transactions and Analytics

The historical (going a decade or so back, anyway) pattern is that cloud processing and storage are often used for analytics and AI; on-premise processing and storage are more for transactional systems. Splitting up analytics and transactional systems has been a core principle of IT management; analytics have been something of a stepchild, and transactional systems have protected status. Even today, 93% of IT executives in the survey agreed that “Our IT department protects operational systems because of data security and privacy concerns.” 90% agreed that, ”Our IT department protects transactional systems because our service-level agreements (SLAs) are commitments that cannot be broken.” 68% agreed that their “transactional systems have limited capacity and IT does not want to impact performance by running analytic processing.”

Given the importance of analytics, both during a business crisis and at normal times, perhaps it’s time to give analytics protected status as well—at least on the level of transaction systems, if not above them. Perhaps analytics should also be generated, and the data for them stored, at least in part on premise-based systems, particularly when there is a need for low latency. Modern on-premise systems can support a variety of different workloads, including advanced analytics and AI. I see no good reason why the on-premise systems used for transactions can’t also be used for analytics and AI.

It seems to me time to move beyond the cloud religion or the on-premise religion, and to choose storage and processing locations that are best suited to a specific use case. In my last and final post on this topic next week, I’ll discuss an approach that makes sense to me.