Its good news when distributed computing platforms let you allocate resources as easily as a teenager manages the playlist on an iPod. Its great news when that distributed solution gets important improvements to scalability, manageability and compatibility with heterogeneous development environments. Thats what were seeing on April 10 from Gigaspaces with its release of its Version 5 infrastructure product line, which commands my attention as one of the best available demonstrations of the ideas in JavaSpaces—and the underlying “tuple space” concepts that are often associated with Yales Linda project but have also evolved beyond that work.
It seems as if theres no such thing as too much information on the system virtualization options available to eWEEK readers, and were trying to meet that demand with coverage from our teams in both News and Labs. It seems to me that the demand for virtualization news is driven by the trend toward enterprise IT workloads with rising peak-to-average ratios. Putting more of an enterprise portfolio of data and supply-chain processing into externally facing systems creates a growing need for systems to respond on other peoples schedules, moving ever farther away from anything resembling optimal batch scheduling of periodic tasks. Conventional provisioning for peak loads becomes ever more uneconomic as those peaks become a higher multiple of the average.
Theres a real cost to the enterprise in offering immediate response to all comers. Users react badly to explicit multitier pricing schemes that make it more expensive for them to get instant gratification, but theres no reason why enterprises cant subtly steer outside users toward less costly off-peak processing —by offering premium content, for example, in response to e-mailed requests thats not available by phone or by online on-demand channels.
We compare “utility computing” to the electric grid, or use terms like “Webtone” to compare future net-based offerings to the telephone grid, but we do well to remember that those other models of transparency fail badly under peak loads. Try to get full voltage in the middle of summer when air conditioning workloads force “brownouts”; try to get a dial tone in the first five minutes after a power outage turns off the lights. Peak loads are expensive, no matter how transparent a virtualization technology may become, and its a good idea to look at ways of leveling those peaks even as we also seek to minimize their costs.
Tell me what youre finding a peak experience at [email protected].
Check out eWEEK.coms for the latest news, reviews and analysis in programming environments and developer tools.