Analysis: Cloud/Virtualization about the ‘right’ capacity, at the ‘right’ time, for the ‘right’ reasons

Following my blog post entitled “Cisco’s latest data center evolution — just more hype?” I received some interesting feedback about how realistic the claims for “dynamic infrastructure” and “on-demand/real-time computing” really are for North American operators using virtualized infrastructures for their own purposes or as a basis for cloud offerings to enterprise customers.

One of the more opinionated responses came from Keith Hanna, head of communications sector, TeamQuest, a capacity management, planning and modeling consultant and provider. “A true real-time environment—whether virtualized or not—I’m yet to see it,” said Hanna, who has plenty of opportunity working with large North American telecom providers and large clients like Fortune 100 companies. “Most of the CIOs, VPs of IT, and data center managers want a ‘dynamic infrastructure,’ but know they cannot expect ‘primetime’ for another three to five years out,” said Hanna.

But with so much talk of this “panacea” in the business, trade and mainstream press every day, what is the hold up?

“There’s not enough focus on the supporting processes needed to build the type of ‘business-aware’ intelligence needed to feed into ‘auto-provisioning’ of virtual machines, or for the deployment and movement of mission-critical applications to virtualization. That just hasn’t happened,” contended Hanna.

“Supporting processes” can mean many things, depending on the scenario. If, for example, a major wireless carrier has to plan for year-end peaks in consumption—say from Thanksgiving through to New Year’s Day—there’s a need for forecasts of demand and analysis of just where and when to automatically drive the “right amount” of compute capacity to support peak loads and demands—and without having to overspend for what will be “idle” inventory most of the time. The goal should be to plan ahead to buy the ‘right capacity at the right time for the right reasons,’ Hanna said.

Another scenario of evaluating “supporting processes” could be billing, where customer sign-up, authentication and billing and payments happen through Web applications that can get bogged down as more and more interactions take place through application environments. To ensure compute capacity in data centers accommodates demands at certain times (say when a new iPad or iPhone, or Android is released), IT needs automated flow-through from business planning and forecasting to execution at peak times of demand. Truly “real-time” application workload deployment and management without human intervention cannot happen unless the vendor community (e.g., Oracle, HP, IBM) or perhaps standards bodies come up with best practice templates to outline best practices and address specifically the processes that ultimately cause lock-ups of systems, where users cannot access Web pages or applications.

Analytics and intelligence to predict problems will rely on understanding “queuing theories and scenarios,” said Hanna, explaining that no application or server environment performs linearly. “Rather they get into queuing scenarios that ultimately get saturated and slowed down enough that Web pages or applications don’t work.”

What happens is that in virtualized environments, applications are put on virtual machines (known as ‘guests’) running on physical assets (servers). Those applications are the workloads and groupings of applications and processes—whether homegrown or COTS—that eventually can clog up or bog down the queuing scenarios and cause alarms that cause systems to drop packets or events, causing sharp downgrades in performance.

“What’s needed is a combination of process, tools and analytics to predict when those queuing scenarios will happen, so that IT isn’t blindsided with lock ups of applications and systems, which happens regularly, even in virtualized environments,” said Hanna.

Where there are many vendors now entering the market on the monitoring side, few provide analytics to predict when systems will start queuing at the physical/server layer or when interaction between workloads and guest operating systems (the virtual machines) will be compromised.

That can be a problem if telcos increasingly offer cloud apps through infrastructure-as-a-service or compute-as-a-service or compute-on-demand scenarios—all of which are often virtualized environments with service level agreements (SLAs) attached. If operators fail to deliver the expected performance, it is their names and brands at stake with large, small- and mid-sized companies and enterprise customers. “If operators engage in due diligence, and ask hard questions of their vendors around virtualization and real-time infrastructure, they can better assess whether they will have the compute capacity on demand that they need to adhere to their customers’ demands,” said Hanna.

He notes if service providers go a step beyond and figure out how to offer the ability to enterprise customers to “peer into the cloud” through portals that offer performance measurements monitoring, reporting and predictions through reports and analysis, that could be an extra revenue generator in the future.

To get to that point, operators need to seek out vendors with experience working with large IT shops and large-scale data centers, as well as develop key performance indicators (KPIs) at both the server and app level to understand performance levels and to predict performance based on historical data and performance statistics.