With great compute power, as Peter Parker’s Uncle Ben might remind us, comes great responsibility — along with a great many things to worry about and items to keep an eye on.
Traditionally, making sure hundreds or thousands of servers and hundreds or thousands of devices attached to them were delivering as needed required big, proprietary, commercial software, like IBM’s Tivoli or HP OpenView. Those monitoring systems work for many organizations, but for many others they’re too stodgy, too bulky and too expensive. For the latter group, it seems there is an open source solution for just about every problem, and managing big distributed systems is no exception. Several commercial companies, including Hyperic, Nagios and Zenoss, have developed network-monitoring tools around open source components.
GroundWork Open Source has been doing it since 2004, and probably has the most extensive system out there for large networks, on par capability-wise with the packages from the large, commercial companies. In fact, IDC analyst Tracy Corbo describes GroundWork’s offering as a “framework for system management” rather than just a performance-monitoring program.
“We use an amalgamated model to build a management solution, leveraging some of the great open source projects out there,” says David Dennis, senior director of product marketing. “There are lots of existing monitors, but they don’t have everything you need, especially for those who have to manage high-performance, high-demand situations. We pull in the tools that will collect the data you need, then we do the coding to make the glue bits.”
GroundWork provides its product in three sizes: Community (freely downloadable; fully functional, but best for about 100 devices; single location); Professional (higher volume; single site or department; adds profiles, alerts, more operating systems); and then the Enterprise version, designed to scale up and monitor distributed environments with thousands of nodes. (“Our largest installation has more than 10,000 nodes,” Dennis says.)
“The Enterprise level adds the technology pieces to deal with a distributed environment,” he says. “This is for people who need to provide high-performance resources on demand, who need to maintain high availability. We add standby servers and child servers to guarantee availability throughout. We provide visibility into the whole grid or cluster with a real-time console and event management, and managers can be alerted by phone, e-mail, or other systems if performance isn’t meeting demand or service levels aren’t being met. The manager can decide then to make changes before problems like latency affect application performance or the user’s experience.”
GroundWork “has ramped up its scalability to support tens of thousands of devices, in a consolidated view with network discovery, traffic graphing, protocol analysis, child servers, distributed monitoring agents, and standby servers,” says Jay Lyman, an analyst covering enterprise software for The 451 Group. “Its Enterprise edition features modular design for distribution of business logic and databases, as well as distributed configuration support. There may still be some limitations to GroundWork for larger enterprise environments. All of the open source players started out mainly on monitoring and aimed more at the mid-market. As they have grown and acquired larger customers, they have grown their software and helped boost the enterprise readiness of open source components.”
Scaling into Grids with Ganglia
Probably the best example of GroundWork “pulling in” other capabilities is its tie-in to Ganglia, the open source monitoring system developed specifically for large distributed environments like clusters and grids. (Ganglia is used by serious customers: NASA; the National Institutes of Health; San Diego Supercomputing Center; Boeing; Lockheed-Martin; HP; Dell; Microsoft; Cisco; and Sun, where it’s part of the Grid Infrastructure reference architecture. You can see it in action at http://ganglia.wikimedia.org/.)
“We bring in Ganglia at the instrumentation and data-collection level to grab detailed information at each node,” Dennis says. “Ganglia has daemons that can be put on cluster nodes to gather metrics, such as CPU utilization, but also more unique things. But it doesn’t have a report engine, or much in the way of alerts.” The Ganglia-to-GroundWork Integration Module feeds the data stream from Ganglia into GroundWork’s dashboards, management consoles, analysis and reporting tools, and notification system.
Ganglia was designed from scratch to be scalable. GroundWork takes advantage of that “to extend Monitor’s scalability as the number of nodes increases,” Dennis says. Monitor’s view, then, scales out to cover every node in the system. All that gathered data can then be analyzed and stacked against customer-defined metrics built into the GroundWork system.
In terms of functions and features, GroundWork can match the “big-name solutions” like OpenView, Tivoli, BMC and CA, Dennis says, “and our scalability is equal to what they offer.”
Where open source, in general, and GroundWork, in particular, benefits really shine is “superior extensibility,” he says. “High-performance enterprises have very individual requirements today. A company will have a critical platform they’ve built on top of Linux, and they’ll have important metrics specific to their application. Not something general like CPU, but something they’ve instrumented that they need to pay attention to. For example, we have a customer that streams video to mobile phones. They have an application running on a cluster, and they have their own set of things they need to measure, like video latency. You can modify an open source solution like ours to take these kinds of custom metrics into account.”
“If you tried to do this with, say, OpenView, well, you couldn’t,” Dennis says. “You can only modify what the API allows you to modify. You can’t readily see what the code is doing, so you’d have to hire a specialist if it could be done at all, or hope HP addresses these issues.”
“High-performance computing started in academic and military research, where you have a whole different set of requirements than in, say, financial services. Usually there’s no revenue on the line,” Dennis says. “But when business moved into grid and cluster computing, and now with the cloud and SaaS, you’ve got agreements between you and the customer coming into play. Uptime has direct financial implications. So you need an offering that lets you run your grid, your IT environment, according to your business requirements.”
Bottom Line
Of course, you can’t talk about open source solutions without mentioning the thing that made open source famous: cost. Annual subscription and consulting fees with GroundWork average about one-tenth that of the big-name brands, the company says. For those seeking all the management and reporting features, redundancy, consulting, and support contracts needed in a mission-critical enterprise environment, GroundWork will cost about 20 percent less than OpenView, Dennis says. Ingenuity Systems’ Farley says their cost was “many times less than a low-end proprietary package.”
Cost alone might not be enough to convince some organizations to implement large open source solutions, but attitudes might start changing when the fiscal hawks start looking at software fees. Software-related costs consume about 30 percent of IT budgets, and 30 percent of that figure is just for maintenance fees, says Ray Wang at Forrester Research. With maintenance fees running 20-25 percent of license fees, some people might wonder if that money could be spent on something else, or not spent at all.
“We really are seeing open source becoming increasingly pervasive in nearly all segments of enterprise IT,” The 451 Group’s Lyman says. “This includes grid, utility and cloud environments. Much of this has to do with the common use of Linux for clustering, but increasingly it has to do with the flexibility, economy and quality of smaller open source tools.”
“There may be a class of the largest enterprise client that would run into scalability issues with some open source system management and monitoring tools,” he adds. “However, few organizations run only one OS, and few, particularly larger ones, run only one systems management software. For those unsure or unwary of open source, but still interested in potential savings or flexibility from open source, I would advise them to limit open source use to divisional, test or development/deployment and go from there.”
According to Dennis, GroundWork’s goods and services help companies “focus on what they’re doing instead of developing a network management and monitoring system.” “We can customize the whole thing around performance requirements, service agreements and very particular metrics,” he says. “If you’re running a gene sequencing grid, you need a system you can tailor to the business of running a gene sequencing grid.”