Picture a Chinese farmer balancing two big baskets full of eggs at each end of a slender bamboo pole as he rides his rickety old bicycle to the next village market. Every so often he hits a bump, and a few eggs fall out, causing him to dismount, rearrange his load, climb back on to the saddle and pedal on. Not exactly the most efficient way to get your eggs to market.
Not the most efficient way to manage a data centre, either. Especially if you have to juggle millions of eggs in hundreds or thousands of “baskets,” or servers, and still hope to avoid winding up with an omelette – and egg on your face.
Increasingly, in modern data centres, servers are not so often viewed as physical boxes or machines. Instead, they are considered to be parts of one huge virtual machine in which resources are allocated, switched around, reconfigured and brought online within nanoseconds, as workloads shift and the demands of business change, sometimes within minutes or even seconds.
This new way of viewing servers and entire networks is called virtualisation, and it is of course a neat way of saving money. It also helps to make the manager’s job easier. “We’re no longer bound by the physical,” says David Grant, data centre manager at Mitel, a leading provider of IP-based voice, video and data services over a single broadband network. Mitel runs its core systems – SAP R/3, an Oracle® database, as well as various R&D applications – from a virtual server environment at its corporate headquarters in Ottowa, but interacts with satellite offices, partners and resellers as well as customers around the world.
By consolidating and virtualising its key IT assets, Grant has effectively done away with things like procurement, installation and commissioning, as well as acquisition costs and lead time. “We simply allocate a new CPU or RAM from the pool of available resources to support new business requirements,” he says. “The environment captures spare resources and redistributes them to where they’re needed most, so we don’t have services sitting idle.”
Automation is obviously key to this kind of approach. “Virtualisation, while useful by itself, must be integrated with management and automation,” says Jacob van Ewyk of HP’s Business Critical Server & Virtual Server Environment Solutions department. He believes that implementing a virtualisation solution without automation in a complex IT environment can leave you worse off than you would have been if you had just left well enough alone.
Sizing systems in data centres is important since that’s what determines how much of IT budget will be tied up in hardware. Therefore, getting an accurate view of real usage patterns is essential. Studies by HP Labs have shown that workloads tend to show lots of sharp spikes, most of them lasting only for relatively short periods, often only a minute or two. The only period of regular sustained usage is usually at night, when batch jobs are running. The rest of the time, demand can fluctuate wildly.
In planning server capacity, for instance for Unix® servers, a good rule of thumb is to aim for about 90 percent of expected maximum demand, since applications tend to slow down perceptibly after that point. That means, however, that an average system will be underutilized for most of the day. In fact, according to IDG, server utilisation commonly runs as low as 15 and 20 percent, which means lots of expensive CPU time goes idle.
In a virtualised server environment, workloads can be combined and allowed to run simultaneously on a single system, thus making much better use of existing CPU resources. But what about those usage peaks? “That is where automation comes in,” van Ewyk insists.
HP Integrity Essentials Global Workload Manager (gWLM), which is part of HP’s Virtual Server Environment, is capable of measuring spikes as they occur, and moving in extra resources within seconds. This fundamentally changes how IT departments size their systems.
Of course, in the past it was always possible to use partitioning to achieve a similar result, essentially moving CPUs around to fit changing demand patterns. However, without automation this meant calling in a systems administrator to do the job, and it could take hours to do what an automated tool like HP Integrity Essentials Global Workload Manager can do in seconds.
The bottom line is that adding automation enables data centres to downsize their hardware and still guarantee performance to meet demands by simply improving CPU utilisation.
Van Ewyk does the maths: "By using fixed-sized partitioning, customers can probably manage with anywhere between 25 and 40 percent less CPU count. Add Global Workload Manager, and you lower your CPU count by up to two thirds, so you’re essentially reducing your server costs by up to 50 percent or more.”
An unexpected result of automation can be increased agility, van Ewyk maintains. If buying a new Unix® server costs a few weeks, then the fact that loading it takes a day is hardly seen as a problem. But if virtualization enables the manager to bring a new server online in a mere couple of hours, then a day becomes a very long time. “It can make poor processes become painfully visible,” says van Ewyk. The result is that the pressure from above tends to start building up to automate these time-consuming manual jobs.
Increasingly, van Ewyk says, companies are turning to virtualisation for missioncritical applications. In the past, the really important systems in a company called for a complete backup system that necessarily would stand idle most of the time, until called upon to step in because of failure in the main system. Now, IT managers can use these backup servers for low-priority jobs, knowing that they can automatically fail over to the standby system and automatically reallocate resources to ensure that the mission-critical workload meets its response time goals. Managers can then save the routine stuff for later when the crisis has passed, thus saving themselves the need for a lot of redundant hardware.
In practice, this means that a company running two data centres, one for its production system and the other as a backup, can now use the second data centre for testing and configuration in addition to backup, thus achieving additional savings.
So in the end it all comes back to a careful balancing act. Only now there really is only a single basket to worry about – one in which you can safely place all of your eggs.
|
Printable version
