Moving to a cloud architecture presents new challenges in data center performance monitoring and management, including...
how to support application performance.
As cloud computing matures, data center systems management is evolving from simply considering the status of virtual machines to application performance monitoring and management (APM), said Jonah Kowall, Gartner Inc. research director for IT operations management. "Many monitoring approaches are still more focused on the infrastructure rather than the app, but that will change," he said.
Get and stay attuned to application owners and their expectations and goals relative to the cloud infrastructure.
IT operations management, Gartner
Application performance monitoring tools should help data center managers determine whether applications are implemented in the optimal manner, he added. Whenever you are dealing with shared resources -- as in a cloud infrastructure -- performance fluctuations in one workload can affect the hardware's other workloads, said Kowall. Application performance is critical and user experience is the most relevant metric of all for data center performance, he noted.
Organizations should aim for an infrastructure monitoring strategy that can transition from "whatever your current state is to something more advanced," he said. Include monitoring subsystems, an event console and APM tools that can monitor multiple technologies. That will help reduce training and other costs associated with expanding monitoring, he said.
Likewise, get and stay attuned to application owners and their expectations and goals relative to the cloud infrastructure, Kowall said.
The trouble with APM
Client expectations and goals are paramount in the shift to virtualization and the cloud at GlassHouse Technologies Inc., an enterprise storage services provider in Southborough, Mass.
"In my experience, we have had only a few clients that actually did in-depth monitoring with third-party tools," said Mike Valuck, GlassHouse practice lead for virtualization. The ones that did, he noted, had a dedicated team that watched and understood the tools "way down into the weeds. They were looking at all the dots and dashes. They are the exceptions." In fact, most companies end up with an off-the-shelf application performance monitoring tool and configure it at its most basic level, where it will send out a simple alert from time to time. The alerts -- and the response to the alerts -- tend to be unsophisticated. "If you want to understand ... seven different kinds of Cisco alerts, that takes specialized people, and most organizations don't have that," he said.
It isn't that IT professionals aren't interested in application performance management, it is a matter of capabilities, time and energy, Valuck emphasized. "People who do more monitoring are getting metrics and reports and are looking at them but (not) in depth," he said. While they may want to know why a database transaction is taking so long, unless they have specialists on board, they won't be able to glean more sophisticated answers. Whatever the monitoring tool does, it does out of the box; they won't understand much beyond that, he said. "That is where we sometimes find clients with five different monitoring tools, and you ask what they are doing with them, or [if they] look at them, they usually just say, 'Well, we get reports,'" Valuck explained.
He sees companies still monitoring and reporting only on capacity issues, "whether it is bandwidth or disk capacity or CPU utilization." That's rational since capacity is usually how they are being charged, with internal infrastructure built to meet peak loads or using a cloud service provider like Amazon. "They pay for a certain number of processors or a certain amount of memory or maybe they are paying for I/O. From a financial perspective, there isn't a lot of incentive to monitor strictly on performance, except in a few cases where it really does come down to a dollar figure of how fast you can get a transaction through, like stock trading speed," he noted.
Most virtualized or cloud-based companies are not attuned to real performance or the end user. One exception to this trend, recalled Ken Copas, cloud services director at GlassHouse, was an airline company that wrote elaborate scripts to allow them go to other airlines' websites and see "what it takes to get an airline ticket." But the focus was external, on the competition.
There are application performance monitoring tools to show where things are hanging up in your own infrastructure, and other tools provide website performance insight. "Those things, if they are used at all, are mostly confined to retail companies," Copas said.
For private cloud or private infrastructure, virtualized or not, the benefits of monitoring and measurement don't really change from the in-the-data-center model, he said. And, of course, if you are operating a private cloud, it is still on-premises, so it really is part of your infrastructure.
More on next-gen monitoring
Learn about server performance monitoring for virtual and cloud infrastructures in part one of this series.
"When you go to a third party or cloud service provider, you either need them to provide that monitoring capability for you, or you need some mechanism to measure inside your provider's environment. And when you go that route, you are taking more of an end-to-end, transactional type of performance monitoring rather than looking at individual performance stats and adding it all up," Copas noted. "I have seen customers that do that, but they are few and far between -- most don't have the mechanism or the plans to actually monitor at that level," he added.
It is easier to collect end-to-end transaction performance data from a cloud infrastructure when it is Web-based than when it is internal. "There are many apps out there that will measure Web performance, because you can just look at the HTML coming into the Web server and measure from first click to last action," he said. By comparison, with an in-house SQL database, for example, "you have to crawl through the transaction logs, which is much more difficult," he added.
Application monitoring for the user's sake
Copas envisions future cloud performance monitoring with a greater focus on the end user. While traditional data centers have tried to implement monitoring technologies for each individual component, it is a best-of-breed approach rather than a unified viewpoint. "Storage tools, backup monitoring tools, network tools -- all of those speeds-and-feeds types of performance monitoring -- are interesting, but not so much in the cloud. ... At the end of the day, users don't care. Only performance monitoring would ever translate into dollars," he said.
Taking the long-term view, Glenn O'Donnell, Forrester Research Inc. analyst, suggests that focusing on monitoring gives too little attention to the factors that make monitoring necessary. "The front end of the process should be better planning and design; APM and other monitoring tools should be the back end," he said.
Unfortunately, planning is more black art than science because it is dominated by the illuminati -- the high priests of technology, who are not always natural planners. "It is an issue of human behavior that crops up in many fields," O'Donnell said.