Browse Prior Art Database

Hierarchical system-level power management policy for high-density web/volume clusters

IP.com Disclosure Number: IPCOM000005126D
Publication Date: 2001-Aug-15
Document File: 6 page(s) / 200K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a hierarchical system-level power management policy for high-density web/volume clusters. Benefits include optimized resource allocation for commodity rack-mounted servers in Internet data centers.

This text was extracted from a Microsoft Word 97 document.
This is the abbreviated version, containing approximately 33% of the total text.

Hierarchical system-level power management policy for high-density web/volume clusters

Disclosed is a hierarchical system-level power management policy for high-density web/volume clusters. Benefits include optimized resource allocation for commodity rack-mounted servers in Internet data centers.

General description

The disclosed policy optimizes resource allocation for rack-mounted servers in an Internet data center based on power usage, performance, and cost. The size of the center can span from a Web farm to a Web cluster. Power resource management extends how the servers, storage, and other resources are brought offline and online, including fine-grain management on each of the servers.

Technology has been converging in the low and high ends of commodity computing. These two convergence points have conventionally been considered disjoint. The low end of the spectrum is the mobile computing platform characterized by the ultra-light and portable notebooks. Optimization is MIPS/Watt optimization with the thermal and power envelope held constant while process and micro-architecture advances include ever-increasing MIPs. Keeping this MIPS/Watt metric in line has also been facilitated by advances in Operating System Power Management (OSPM). The most prevalent example in the horizontal computing industry is Advanced Configuration and Power Management Interface (ACPI).

On the high-end of the spectrum, the optimizations have been in MIPS and I/O throughput. The means by which continuously increasing throughput has been achieved include:

Parallel host bus systems

Fast I/O buses

Larger SMP clusters

Moore's Law

Micro-architectural and process advances from which the mobile computing regime also benefits

Detailed description

For the conventional data center, multiple commodity web server farms content is typically replicated on each server. Additional servers are added to handle the increase in request traffic or to support fail-over for high availability. In a web cluster environment, the web content is typically distributed across the front-end servers. The cluster may also be replicated a number of times with distributed content to handle increase in request traffic or to support fail-over.

Web server farms or clusters can be aggregated into single virtual server by means of a front-end, load-balancing server(s) as shown in Figure 1. This graphic illustrates a front-end cluster of web servers in a data center. The model is a load-balancing configuration that load balancer accepts requests from the external network and dispatches them across an internal, higher-speed switched network.

A conventional alternative is to use a software-based approach for load balancing. One IP address is published to an external network, such as WAN or the Internet, but internally the particular requests are routed to the internal network of servers. This approach enables a one-to-many mapping to balance the load across many entities. The technique is such that the front-end load balance...