ERCIM news 135
ERCIM news 135
ERCIM news 134
ERCIM news 134
ERCIM news 133
ERCIM news 133
ERCIM news 132
ERCIM news 132
ERCIM news 131
ERCIM news 131
ERCIM news 130
ERCIM news 130
Back Issues Online
Back Issues Online

by Lars Rasmusson

What would computing be like if we had a virtually infinite pool of computing power at our fingertips? And how would we go about creating such a pool? These and other related questions are the subject of research at SICS.

SICS is building infrastructure that will provide computing resources on demand, so-called ‘utility computing’ or computing power on tap. The name stems from an analogy made with traditional utilities, such as water and gas. Traditional utility resources are acquired as they are needed, without the need to reserve them in advance, and are paid for in proportion to usage. This is very different from the way in which computing power is acquired. We must buy the machinery that generates the computing power, rather than being able to easily buy the power only.

The next-generation computing infrastructure

Who will benefit from having computing power on tap? The initial beneficiaries will be companies with fluctuating workloads and problems balancing out the workloads over time. As things stand, it is common that, in order to achieve an acceptable quality-of-service at peak load, business servers on average run at only 5 to 10 percent load. Such an amount of unused resources translates into huge costs. These costs could be cut with the help of utility computing, where someone else employs the unused resources when they are not needed by the business.

What's more, having a fixed computing pool can become a straitjacket that stifles innovation of new services and damages the flexibility of a company. For instance, we see more and more businesses that produce services on the Web. These new services are often the most innovative users of the new network and information technology that we see today. Network businesses are vulnerable to flash crowds – hundreds of thousands of users that access an online service at the same time – because it is uneconomical to dimension resources to deal with such intense load peaks. For such businesses to be reliable and viable, we need infrastructure that can allocate resources to these businesses in a short time, from a common resource pool.

Can a virtually infinite resource pool be built? Yes, but it will require an infrastructure that encourages investment in scarce resources. More investment in the right places enables the pool to grow organically, and to adapt to unknown workloads that will occur in the future and services that we cannot yet begin to imagine.

To begin creating an infrastructure that turns computing power into an almost liquid resource, SICS is building, in cooperation with HP Labs, the next-generation computing infrastructure ‘Tycoon’. This provides a public, open market, where computing resources are sold for profit. The market is designed to encourage on-demand resource acquisition, rather than reservations. The motivation is that you rarely know your future demands in such detail that they could be easily expressed in a resource reservation language. Instead, you typically notice at run-time that more resources are needed to meet the quality-of-service objectives for response latency, throughput and so on.

The Tycoon market uses dynamic prices that fluctuate in response to changing supply and demand. Demand-driven prices make it more profitable to provide scarce resources, and encourage resources to be added to the network when a shortage occurs. The market is open, meaning resources are provided by competing providers; this prevents consumers from being locked in by a particular provider.

SICS is also active in the EU-funded SORMA project, led by the University of Karlsruhe. SORMA's goal is to create a truly open and ‘self-organising’ marketplace for the future computing grids. In contrast to Tycoon, which provides one specific computing platform built on virtualization, SORMA will be a general market platform or hub, that will connect together heterogeneous computing platforms in one common infrastructure. SORMA’s members are active in many of the existing Grid computing initiatives worldwide, such as gLite, Mosix, and CATNETS. We expect these will benefit greatly from the market features that will be provided by the SORMA infrastructure.

Link:
http://tycoon.hpl.hp.com/

Please contact:
Lars.Rasmusson
SICS, Sweden
Tel: +46 8 633 15 00
E-mail: Lars.Rasmusson@sics.se

Next issue: January 2024
Special theme:
Large Language Models
Call for the next issue
Get the latest issue to your desktop
RSS Feed