Virtualization and the Power Management Imperative

Virtualization’s exploding popularity is easy to understand given its potential to lower hardware spending, simplify administration and boost availability. Article by Hervé TardyVirtualization’s exploding popularity is easy to understand given its potential to lower hardware spending, simplify administration and boost availability, so it comes as little surprise that nearly 80 percent of server workloads supported by x86 hardware will be running on virtual machines (VMs) by 2016, according to analyst firm Gartner Inc. IT and facility management teams, however, must account for some new power-management realities, because despite all of its advantages, virtualization subjects datacenters to many new, significant power-related challenges. The upside is that solid, affordable solutions not only address these challenges but provide opportunities for businesses to extract the full potential of the electrical savings offered by virtualization and server consolidation, all while ensuring power capacity, data integrity and business continuity are never compromised. This article explores major power-related considerations, and available solutions, associated with implementing server virtualization; it then outlines strategies for preserving business continuity and data integrity in virtualized infrastructures so you can fully maximize the benefits of virtualization. The Powerful Considerations of Virtualization 1. Overall power consumption may be lower, but each server draws more power. Virtualization tends to increase the amount of time a server runs at or near peak performance. On an unvirtualized platform, the average server CPU (central processing unit) runs at only 10 to 15 percent of capacity. With virtualization, that figure jumps to about 70 to 80 percent. The higher the CPU utilization, the higher the power consumption per server. For example, blade servers can normally consume from 3 to 6kW per blade chassis—as much as 8kW in high-performance compute configurations. To solve this problem: Increase the density of enclosure-level power protection and distribution. Available enclosure-based power modules can distribute up to 36kW in only a few U of rack space. Additionally, the newest generation of enclosure-based power distribution units distributes power in an organized manner to between 4 and 45 receptacles for a wide range of power densities. 2. Each enclosure draws more power than ever. Virtualization reduces power consumption overall, but it increases power consumption per U and per enclosure. Traditionally, IT managers could plan for about 60 to 100 watts of power consumption per U (1.75 inches) of rack space; a full rack of equipment averaged 3–4kW of power. With today’s blade servers and VoIP systems, that figure has escalated to 600–1,000 watts per U and keeps growing. Power consumption may soon reach up to 40kW per rack. To solve this problem: Build out a power sub-distribution strategy. Instead of running individual cable drops from your large UPS and PDUs to each rack, run higher-powered subfeed circuits to an intermediate remote power panel (RPP), power distribution rack (PDR) or rack-mounted power distribution device—and from there to enclosures. 3. Higher rack densities may exceed available UPS capacity. In traditional datacenter designs, one or two large, three-phase UPSs stood alone in a separate room, providing conditioned power and battery backup for the whole datacenter, perhaps even the entire building. With virtualization, the capacity of the existing, centralized UPSs can become a bottleneck to computing efforts, especially as redundancy becomes more important, owing to the resulting cluster of high-density racks in a datacenter that was designed for lower-density racks. To solve this problem: Enclosure-based UPS systems with power densities of 2kW or more per U can meet these new high-density computing demands. Modular, scalable UPS systems can be deployed in a variety of system architectures for centralized, zone or distributed power protection, along with the needed redundancy to meet business objectives. If you combine the UPS with a high-powered distribution system, you can deliver power to loads of various voltages, power cords and layouts with flexibility. 4. Workloads can be dynamically reallocated, but the support infrastructure can’t do the same. Virtualization makes it possible to deploy, move or clone an application from one platform to another at any time, even while it’s running. That means energy demands can shift around the datacenter, but power is a fixed asset, tied to a physical infrastructure that can’t be pressed beyond the limits specified by the National Electrical Code. This on-demand migration of applications demands new levels of visibility into how IT applications affect power, and vice versa. Is the IT equipment that runs the most important applications receiving computer-grade power? If you move processing-intensive applications, will you overload a branch circuit? Which rack has enough power to accommodate new virtualized servers and their applications? To solve this problem: Employ a 24/7 power-quality metering, monitoring and management system at the branch-circuit level, which can be conducted at several points in the datacenter power distribution system. The right strategy will be a trade-off between the number of IP addresses you’re willing to allocate, the number of devices to be monitored and the degree of detail required. Preserving Business Continuity and Data Integrity Many organizations use virtualization management software to administer host servers, VMs and more, but users of virtualization management suites must usually employ a separate set of management tools to monitor their power infrastructure, decreasing the productivity of their technicians and potentially delaying response times when problems occur. Deploying modern integrated power-management software solutions significantly eases the complexities of keeping critical applications continuously available during power outages. For starters, some such solutions integrate directly into the dashboards of leading virtualization management products, including VMware vCenter Server and Citrix XenCenter, among others. This integration enables technicians to view, monitor and administer not only physical and virtual servers but uninterruptible power systems (UPSs) and other power devices through a single console. Avoiding data loss and keeping critical workloads operational are top datacenter priorities during power outages. Preventing data corruption is essential as well, and sometimes shutting down servers is the only way to meet that goal. To do so, datacenters have long relied on a combination of UPSs and power-protection software. Server virtualization, however, makes safeguarding data during power failures significantly more complicated. Most often, when a virtualized datacenter loses power, technicians must manually shut down not only their physical servers but the virtual machines running on those host servers as well. Additionally, they must execute the many steps in that process in a specific sequence, often in the face of intense time pressure. Companies can overcome these challenges in several ways. Download open-source management code: Pre-written, open-source operating system code for shutting down servers gracefully and in the correct sequence during power failures is being developed and distributed via websites such as Network UPS Tools (www.networkupstools.org). By downloading, installing and customizing such code, datacenter managers can equip their infrastructure to shut down servers in the proper order when utility/server power becomes unavailable. This solution offers an easy yet extremely powerful and highly configurable power-management option to organizations that use and customize Linux or other open-source solutions—a category that includes most operators of cloud computing datacenters. Add automated scripts to advanced power-protection software: Many advanced power-protection solutions enable users to create scripts that automatically respond to specific alarms in a predefined manner. Companies can use such scripts to augment their power-protection system’s built-in functions in sophisticated ways. For example, technicians could extend UPS battery runtime by creating a script that automatically shuts down virtual machines running non-critical workloads early in a power outage and then consolidates the remaining virtual machines onto a smaller number of host servers. Deploy advanced power-protection and -management software: Though power-protection applications enable organizations to shut down servers in an orderly manner during utility failures, most such systems support physical devices only. The latest and most sophisticated solutions, however, support virtual machines, as well as hosts such as VMware ESX, Microsoft Hyper-V, Citrix XenServer and Red Hat KVM. Consequently, they can be configured with the aid of the virtualization management system to shut down both physical and virtual servers in pre-defined sequences that minimize exposure to data loss. The latest solutions can also enhance productivity and response time while protecting data integrity by identifying power failures immediately, initiating a backup site, and triggering disaster-recovery failover and planned migration processes, transferring data safely when shutting down servers is unavoidable. Closing Thoughts The flexibility that helps make cloud computing and virtualization so attractive can also inadvertently produce downtime, so a truly comprehensive power-protection environment must be designed to ensure that essential applications remain continuously available. Application demands affect power consumption, and power consumption affects application availability and performance—yet rarely do these two interdependent groups work in unison. This reality calls for a monitoring and management system that merges IT and facilities under a single pane of glass—one portal to view the status of IT resources, power and thermal conditions. With this type of holistic perspective, IT and facilities managers can easily see when abnormal conditions threaten the IT infrastructure—and exactly which business services are at risk—so proactive measures can be taken. Modern power-monitoring and -management solutions position companies to take full advantage of server virtualization’s rewards, but most importantly, modern solutions deliver confidence that power systems are doing what they should, that data integrity will never be compromised (even during power outages) and that personnel will be immediately notified of alert conditions in time to resolve them, not just react. These solutions empower users to predict, manage and prevent power problems before they even have a chance to touch the bottom line.