Dr. Dobb's Journal - March 2008 - (Page 10)

Alia Vox by Rakesh Kumar Economic Models for Many-Core Processors Mix and match. Will many-core CPUs go the way of cell phones? IMAGINE WORKING ON your computer and suddenly realizing that you need more processing power than you currently have. Now consider the ability to simply go online to the computer or processor vendor’s website, submit your credit-card information, specify how much more processing power you need, and for how long, hit a button, and, bam!—there appears a pop-up that says, “Congratulations, your hardware has been upgraded.” Similarly, imagine the ability to downgrade at will or, for that matter, pay for processing power just like electricity or cell phone service—only for what you have used. We believe that such economic models The last two models let users treat processing as a service by may indeed become feasible for many- decoupling payment from ownership. These models are in addicore processors. Current economic mod- tion to an Intelligent Baseline model, where users make one-time els for processors require customers to decisions about the number of cores they need, and vendors estimate their average-case or worst-case appropriately enable a subset of cores on the chip before shipping. computational requirements beforehand, There is also a strong economic motive to look at alternative ecoand then buy a processor or system that nomic models. Currently, each chip has exactly one peak computameets those needs. This causes computa- tional capability. Because the number of unique chips is decreasing tional resources on chips to often not due to prohibitive manufacturing costs and diminishing profit marmatch computational requirements of gins, there is an increasing disparity between the computational customers and their applications. So power that a customer needs and the capability that a chip can proprocessors are often under- or over- vide. This again results in wasted dollars and reduced efficiency. The utilized, both resulting in “wasted” dollars. problem gets worse with increasing number of cores on a die. And the problem gets exacerbated as the Economic models such as this are not new. number of cores on a processor increase. Large-scale server systems (or mainframes) also use a CapacityJoseph Sloan and I have published a On-Demand (COD) system where the economic models are simitechnical report (www.crhc.uiuc.edu/ lar—the IBM COD systems being just one example. The primary dif~rakeshk/techrep_economic.pdfm) that ference is the nature of customers. For the COD systems, the cuspresents four alternative economic mod- tomers are typically a relatively small number of trustworthy, visible els for many-core computing. The pro- companies that can be tracked and made accountable if they do not posed models recognize that when a subscribe fairly to the economic model. With many-core procesmany-core chip is bought, the customer sors, authentication becomes a concern due sheer volume and may often wish to pay for fewer cores anonymity of customers. Similarly, security and robustness become than what is present on a chip. The mod- a concern as anonymity may encourage malicious users/customers els provide the customer the flexibility to to unlock the processing power. Also, reversibility (or downgrades) change the available peak computational needs to be enforced more stringently as there are fewer ways to capability of the chip (or effective number force the customer to give up resources. This difference in the of cores) during the lifetime of the chip. nature of customers translates into unique support and enforcement mechanisms. In fact, the bulk of our work in this area is devot• The Upgrades Only model lets users ed to looking at the hardware and system-level mechanisms that are adapt to increased average compuneeded to enable the economic models. Mechanisms range from tation needs by scaling up any time techniques to uniquely identify a chip, to securely authenticating a postfabrication and postpurchase. • The Limited Up/Downgrade model user, to locking access to unauthorized cores. Most policing mechalets users adapt to changes in comnisms need to be implemented in hardware and made inaccessible putational need by upgrading/downto the user to ensure the integrity of the models. grading the chip on demand. Several other issues need to be worked out to make these models • The Cores On Rent model lets users rent access to cores as long as the harda reality, the biggest being the concern about the licensing model for ware has a valid lease. Access to cores software. Conventional software licensing models tie licensing to a is revoked upon lease expiration. peak computational capability. The proposed models allow changes • Finally, the Pay Per Use model measin the available peak computational capability of a chip and may, ures and records hardware usage therefore, require unconventional software licensing models. and the billing happens over a specified lease period. But then again, imagine… DDJ Rakesh Kumar is an Assistant Professor in the Electrical and Computer Engineering Department at the University of Illinois at Urbana Champaign. He can be contacted at passat.crhc.uiuc.edu/rakeshk/. 10 There is a strong economic motive to look at alternative economic models Dr. Dobb’s Journal l www.ddj.com l March 2008 http://passat.crhc.uiuc.edu/rakeshk/ http://www.crhc.uiuc.edu/~rakeshk/techrep_economic.pdfm http://www.crhc.uiuc.edu/~rakeshk/techrep_economic.pdfm http://www.ddj.com

Table of Contents Feed for the Digital Edition of Dr. Dobb's Journal - March 2008

Dr. Dobb's Journal - March 2008 Contents Hmmmm Alia Vox Developer Diaries Developer’s Notebook Social Networks and Software Development Conversations Detecting Bugs in Safety-Critical Code Change Code Without Fear Continuous Integration and Performance Testing Wt: A Web Toolkit Automating Release Notifications The Agile Edge Effective Concurrency Swaine’s Flames

Dr. Dobb's Journal - March 2008

For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.