Big cloud hurdle
Workload readiness is the key to accelerating cloud adoption
Business executives are buying into the promise of cloud computing. And there are sufficient examples of “ready to go” compute workloads— mostly data storage and retrieval based on this survey—to create the impression that a wholesale shift to cloud is upon us. But not so fast.
For various technical reasons, most missioncritical workloads are not yet ready for multitenant cloud computing infrastructure. And they won’t be ready until some major re-architecting and re-coding gets done, the cost of which may prove to be a major factor in keeping these applications in traditional data centers for many years.
Mission-critical online transaction processing (OLTP) applications and other enterprise systems will require extensive recoding to take full advantage of the highly virtualized cloud infrastructure. IT outsourcing (ITO) providers, Infrastructure-as-a-Service (IaaS) providers and enterprise cloud adopters are unlikely to see cloud technology fully replace traditional infrastructure in some sort of “forklift upgrade” as a result. And once the commercial applications are ready, large organizations aren’t likely to move them to the cloud until the refresh cycle dictates it based on the business requirements of individual enterprises.
Findings from the FORFIRM IT Outsourcing and Cloud Computing Survey of 489 CIOs and other senior executives paint a clear picture of the situation:
• Only 18 percent of the respondents said 51 percent or more of their organizations’ workloads were ready for the cloud today; 53 percent of respondents said 25 percent or less of workloads were cloud ready. (See Figure 1)
• Three years out, the figures were essentially reversed: 56 percent of respondents said 51 percent or more of their workloads would be cloud ready; 20 percent said 25 percent or less would be ready.
• Data storage and retrieval was the most widespread application running in the cloud today, followed by OLTP. (See Figure 2) In three years, the frequency of each increases and the order does not change.
ITO and IaaS executives, who reviewed the survey findings, wondered if the respondents weren’t too optimistic, especially about OLTP workloads. “Naive,” was how one executive described expectations for an easy transition of OLTP to the cloud—at least for mission-critical workloads.
At the heart of most of these applications are one or more relational database management systems (RDBMS). This is a problem, executives say, because RDBMs tend not to perform well in a highly virtualized environment. They can obtain significant performance gains when run on a cluster of tightly-coupled servers; many IT shops in financial services, for instance, have aggressively embraced that concept. In the highly virtualized infrastructure of the cloud, where entire software stacks can be redeployed at a moment’s notice, the coupling of servers is quite loose. In some cases, this causes distributed RDBMS performance to suffer significantly.
“Clouds are notoriously bad at optimizing relational database workloads,” says Mark Bilger, vice president and CTO for Dell Services, which recently began to offer IaaS. “But RDBMS are the workhorse for 99 percent of traditional ITO applications. You get some benefit from moving to cloud, but you cannot massively parallelize these apps as you can a Web 2.0 application.
Mission-critical workloads may not be cloud ready for a long time, But the great volume of computing is not mission critical and could be moved to the cloud.
The good news, of course, is that the mismatch between apps and infrastructure offers a significant business opportunity. Established ITOs and IaaS upstarts stand to profit if they can help enterprise customers re-craft their home-grown apps and adapt commercial software packages for the cloud. The developers of those packages —SAP, Oracle, IBM, Microsoft and others— would also play a potentially major role in this effort.
The opportunity for the ITO providers is to partner with applications providers in refactoring and rewriting ITO applications. But fairly exotic software technologies, such as Hadoop and BigTable, are needed to fully exploit the cloud, and I don’t think that the majority of CIOs understand that.
Evolving database technologies should also enable OLTP to run well in the cloud. They will need two or three more years to mature, which means that it will be six or seven years before they are widely adopted.
Not surprisingly, the survey found security to be the top perceived risk of the public cloud, but “data and system portability,” which is related to migration, was tied for second.
Not all IT activity is affected equally by security. It’s a shame that more Web-serving workloads aren’t in the cloud. They are, by definition, public-facing, so there’s no compliance or security issues there, and they’d get much better performance, much better cost, much better scalability using public servers.
In the survey, web-serving workloads were next to last in frequency in the cloud, followed only by analytics workloads, which ITO and IaaS executives see as another missed opportunity.
These are the workloads we recommend as the low-hanging fruit. The most value and the biggest growth are going to be in the analytics space, because you have all that high-performance computing that you can take advantage of without all the expense.
Analyzing the application portfolio is very important. Some applications will never move to the cloud. We find that this assessment initiates some rewriting and combining of legacy apps to take advantage of cloud and new database models like Gemfire or technologies like Hadoop. For the rest of the workloads, there are highly automated and industrialized refactoring services that quickly provide immediate return on investment.
The operating system is yet another obstacle. Most computing clouds today are built around Linux and Windows, and sometimes within enterprise IT organizations you still have apps built on servers and operating environments that pre-date virtualization and cloud computing. Those apps don’t really lend themselves nicely to moving to a public cloud or even a private cloud today in a virtualized fashion.
Apps like traditional corporate email systems tend to be very scale-up as opposed to scale-out. When you want to add capacity, you buy a bigger server.
In the cloud, to boost performance you add more servers, not bigger ones. People are rethinking how they’re going to build apps to take advantage of the cloud’s scale-out architecture. Packaged apps are going through this transition.
Many organizations have decided to put mail in the cloud. But despite financial advantages in many cases, even moving this workload can be more difficult than some perceive.
There are very few enterprises, especially larger ones, for whom mail is just mail. In many cases, mail is tightly integrated with collaboration services and workflows, and it involves sign-offs and authorizations around critical applications. This creates certain challenges. Boles says software providers are starting to address these challenges
The most important step for many companies is to finally—really— get serious about enterprise architecture and its role as a planning tool.
There are many great reasons to consider cloud infrastructure, and it is definitely the future. But the transition is not free. Before re-hosting major parts of the application portfolio onto clouds, an enterprise would be wise to revisit the strategic fit between the evolving business architecture and IT. It’s also a great time to commit to disciplines such as ITSM (IT service management) because cloud infrastructure fits naturally into the ITSM paradigm.
Despite efforts directed at workload readiness, it’s going to be very difficult, time-consuming and expensive to modernize and upgrade legacy applications, and that’s going to take some time. You will see traditional data centers continue for many years.