Guidelines for Green Computing

Traditionally, electrical power usage has not been considered a key design criterion for datacenters and computers, nor has electrical consumption been effectively managed as an IT expense. Today, the most famous industry leaders, the most prominent industry analysts, the largest government agencies, and the most diverse environmental groups offer a new and collective viewpoint: that energy efficiency is the most significant issue facing technology providers and their customers today. This is due to exponential increases in power and cooling costs over the past few years, and to customer demand for concentrated computing that is outpacing the availability of clean, reliable power in numerous locations, worldwide. Today, computers account for approximately 2% of worldwide energy usage. According to Gartner analysts, by the end of 2008, half of the world's datacenters will not have enough energy capacity to meet the power and cooling requirements of the latest high-density computing equipment, such as blade servers.[1]  Consider the fact that it is now possible to pack racks with computing equipment requiring 30,000 watts-per-rack in connected load, compared to 2,000 to 3,000 watts-per-rack a few years ago. Gartner estimates that energy bills, which traditionally have accounted for less than 10% of an overall IT budget, may soon account for more than half.[2]In the U.S., for example, consumption of electricity by IT has doubled since 2000, and now comprises 3% of total electricity consumed nationally. 

Electrical power for datacenter servers is only part of the problem. Non-IT devices also consume datacenter power, including transformers, uninterruptible power supplies, power wiring, fans, air conditioners, pumps, humidifiers, and lighting. Almost all of the electrical power feeding the datacenter ultimately ends up as heat, and traditional datacenters typically waste more than 60% of the energy they use to cool equipment, according to Gartner.[3]In the case of PCs, about half of the power consumed is wasted.

Reducing the Carbon Footprint

Because the vast majority of hardware devices have been designed to provide maximum functionality and performance with little regard for wider environmental issues, according to The Green Grid Consortium, Gartner's analysts advise enterprises to make the most efficient use of existing systems in their datacenters, using technologies such as workload balancing and virtualization, and to plan for greener datacenter designs in the coming years.[4]Both Microsoft and Yahoo are establishing server farms in eastern Washington State, near less costly hydroelectric power supplies, and Google now has a datacenter near hydroelectric power supplies in Oregon. The Climate Savers Computing Initiative Group estimates that if its consumption targets for PCs and server systems were met, the amount of carbon dioxide emitted to supply power for computing uses could drop by 54 million tons a year and save more than $5.5 billion in annual energy costs. The group also targets increased efficiency for voltage regulators, and recommends that organizations use power-management software on their PCs, which puts desktop computers into sleep mode to conserve energy during periods of inactivity.

ADI and the Green Datacenter

The key to any effective response to these challenges is to be able to maintain, and even increase, the capacity of IT infrastructure while reducing its energy consumption. An ADI includes many of the components required to enable IT groups to achieve green objectives through consolidating or offloading server workloads in the datacenter, and reducing energy consumption at the endpoint. Centralizing all applications and data for delivery to users can have a positive impact on datacenter power consumption. Intelligent server provisioning and load management can ensure that only the hardware needed to meet user requirements is actually being used. For example:

• Server provisioning can be used to reduce the power consumption of servers by up to 15%, by streaming the entire operating system to diskless server hardware. Further, its bare-metal provisioning capabilities can allow for time-based server configurations, such as application virtualization and delivery during business hours, and back-end processing operations at night.
• TCP optimizations, traffic management, caching, and SSL acceleration capabilities that efficiently offload Web server functions are already being used to reduce the amount of hardware required to serve many Web platforms.
• Reduced power consumption can be extended to the office, by replacing traditional desktops with both energy-efficient thin clients, and application- and desktop-virtualization and delivery.

Beyond Energy Efficiency - Taking a Proactive Approach to Green IT

Although the first steps will likely be directed toward more efficient use of power in the datacenter, green IT initiatives must also take into consideration the indirect positive and negative impacts of energy usage to other aspects of application delivery. Taking a positive approach to finding these impacted areas brings about opportunities to widen the scope and focus on the end-users. In the office, server-side app virtualization can extend the life of corporate desktops, which often contain hazardous and non-biodegrading materials, thereby reducing both the frequency of desktop-refreshes and the amount of discarded equipment. IT can become an enabler of more sustainable business and even personal work practices, by focusing on carbon emissions from travel related to business, such as commuting to the workplace and to meet customers, business partners, and colleagues. Among the key functions of ADI is to provide a secure single point of access to all applications that automatically customizes access policies based on user scenario. In addition, applications and desktops can be virtualized and delivered to users working from home. Further, users can securely work online on their business desktops from home and other remote locations, and geographically dispersed teams can securely use online collaboration tools to conduct effective meetings without the need to travel. These capabilities can have a significant effect on vehicle carbon emissions, because they reduce both the number of vehicles and the number of people traveling at the same time. Traffic flows more freely and less fuel is burned.

Implementing a Green ADI

Geographically distributed systems can lead to unnecessary duplication; however, centralizing an organization's IT infrastructure into one or more datacenters has the potential to lead to power and heat densities. There is a tendency to over-provision hardware when planning capacities for an application or service. This tendency is understandable because, to ensure minimum availability - whether tied to an SLAor informal user expectations, you need to scale for peak expected loads. The result is that most of the time, user load is either balanced across all of the servers in the silo, or unused servers are still running in anticipation of increased connections. This, in turn, results in a far greater consumption of power than was actually required to support the business.

The ADI contains virtualization technologies that can be used to remove the hard-coded links between the different elements that make up an application platform -network, hardware, operating systems, applications, and user settings. When applied, these technologies create a flexible computing infrastructure that uses only the hardware that is actually needed.

Reducing Power Consumption at the Datacenter

System virtualization and provisioning services can be used to create the smallest practical number of server operating system images that would be used only when required. System virtualization is already widely recognized as a significant enabling technology for consolidating a large number of servers into a smaller number. If the consolidation effort is a transition from using legacy hardware platforms to using new energy-efficient platforms, there is an additional reduction in total power consumption. A lesser known technique is to provision multiple homogenous servers from a single, centrally stored operating system instance using system streaming. The ability to re-factor a server from running one application or function to another, as needed, can reduce the number of servers required, especially if loads are not consistent during certain times of operation. System streaming can also remove the need for local hard drives, reducing power consumption per server by approximately 15%. When hardware cannot be shared between functions, this dynamic provisioning technique can be used to quickly bring servers online from a switched-off state. When servers are not being used, they are not consuming power. A useful technique to ensure that there is always a minimum number of servers supporting an application is to separate hardware into two rings. Ring 0 contains the servers that are always running, keeping a separate Ring 1 of servers that can be added or removed from the service, as needed. Another technique is to extend load balancing systems to automatically provision servers on demand. Citrix's Netscaler Web Application Controller uses virtual local area networks (VLANs) that contain a pool of IP addresses tied to an application, such as SAP or an eBusiness Web site. Each VLAN can have a threshold set against maximum user connections. Citrix XenApp has load balancing rules which can also be set against user connections. When a user connection threshold is reached, it should trigger the provisioning of additional server images from Ring 1 that boot and then register their IP addresses into the VLAN to begin receiving connections. When the user load dips below the threshold, the Ring 1 servers should stop receiving connections, and they should be powered off as soon as there are no longer any users.

Server Load Balancing Based on Power and Heat Consumption

Monitoring systems tend to focus on utilization for components, such as CPU and memory. Many of today's server platforms have management systems that measure and report on heat and power consumption. These measurements can be used to provide real-time metrics that, if combined with the dynamic provisioning techniques mentioned above, could create an intelligent load balancing system that optimizes load based on the power being consumed. Distribution can occur on a user basis or on a server image basis. For heat, distribution would aim to lower utilization of overheated servers until they cool down. For power, the aim would be to use as few servers as possible without degrading performance, as the difference in power draw between a server with low utilization and a server with high utilization is quite minimal.

Consistent and Transparent App Delivery

By designing an ADI that ensures applications and desktops perform as well when delivered outside of the office as they are when consumed within the office, the perceptual impediments to working at home and remotely can be reduced. Best-practice designs also ensure that the process required to start an application delivery session is identical when users request access to applications in any scenario. This consistent and transparent application delivery process can occur over a range of WAN links, by using a combination of server-side app virtualization (low bandwidth) or client-side app virtualization (high bandwidth) for delivering Windows applications, Web application delivery controllers for delivering Web applications, and WAN optimization controllers for accessing data from locally installed or streamed applications at the endpoint. The appropriate method of application delivery for a particular access scenario can be determined by implementing application access control in conjunction with endpoint scans. Using this method, criterion such as network link performance, endpoint configuration, and available application formats (for example, Web-based alternatives, such as Outlook Web Access for Microsoft Outlook or Netweaver for SAP) are taken into account by the ADI to make the decision on behalf of the user. Implementing thin clients at the endpoint is another method of reducing overall power consumption. Applications, or entire desktops, can be delivered to these thin clients using server-side app virtualization.

[1]"The Green Grid Opportunity - Decreasing Datacenter and Other IT Energy Usage Patterns", The Green Grid, 2007
[2]Ibid.
[3]Ibid.
[4]Ibid.