Add to that the occasional need for extra capacity because of business peaks or special requirements such as systems maintenance work, and you have a capacity planning conundrum. How do you find a happy medium between constantly bumping up against the limits of your computing capacity and paying for capacity long before you need it?The answer is capacity on demand. About two years ago, Shell installed an IBM eServer i890 as its main system – the platform for its PeopleSoft EnterpriseOne business applications suite (now an Oracle Corp. product). The machine has 32 processors, but when Shell installed it, only 24 were turned on.
Over the past couple of years, Shell fired up additional processors, first just temporarily to handle demand peaks. Over time, five of the eight processors that were originally spares have gone into permanent service, so the system is now running 29 processors all the time, according to John Thiers, senior staff systems analyst at Shell in Calgary.
Shell doesn’t pay for processors until it uses them, and the company can activate additional processors quickly with a phone call or e-mail to IBM, Thiers says.
Since Shell had the system for about 18 months before increasing the number of processors in full-time use, Thiers says, the company saved significant amounts compared to what it would have spent had it simply installed a fully activated 32-way system in 2003. And Shell has not had to wait for system upgrades to be arranged before it could add capacity.
“We need the capacity on demand,” Thiers says. “Otherwise, we’d be into upgrading the machine faster than we’d like to.”
The upgrade will still happen eventually. Thiers expects the i890 to reach its limit in the next couple of years, and when it does he expects Shell will replace it with another machine offering capacity on demand. He adds that another server, used for data warehousing, will be replaced with a capacity-on-demand system when it maxes out.
He says one benefit of the on-demand approach is the “cost smoothing” it provides — instead of paying for all the capacity when a machine is installed, Shell can defer some of the cost until it is actually ready to use the added processors.