BETTER backup

Today's complex systems require high availability. That means power systems must step up to the task

The telecom industry is on the upswing. However, its centralized design methods are antiquated compared with today's capabilities. Significant improvements could be realized, even in the central office environment.

Availability, or uptime, measures the ability to provide service to the consumer. Data centers and COs offer the service with two cascaded functional blocks. The end-user equipment provides the desired functionality. A second block addresses backup with either uninterruptible power supplies (UPSs) in data centers or battery plants in switching offices.

In the past, these centralized approaches were the desired choice for some good reasons. The methodology freed the end-product developer from either developing the expertise or taking the responsibility for backup generation. Also, a UPS was the practical backup choice for computing equipment because it fit well into the centralized multiple output rectifier embedded within systems. It is basically an AC-input-to-AC-output device. The consumer may choose to provide backup as desired without affecting the end-user equipment.

Telecommunications opted to provide a -48 VDC battery plant for the CO. It thus offered a backed-up DC source to the equipment. More than two-thirds of the CO load consisted of voice circuits that used the generated -48 VDC directly. It was obviously a beneficial, practical and cost-effective solution for switching equipment in COs.

The right direction The evolution of the Internet and the integration of voice and high-speed data define new performance requirements for mid-level to high-end computing equipment. Co-location of data-crunching computers and voice switches brings expectations of the ability to seamlessly interface into the telco or data environment.

Computer manufacturers face the dilemma of meeting or exceeding the availability achieved by switching equipment. This is not a simple challenge. In addition, rapid deployments are the norm to meet recent market expectations.

One possible way to improve performance, enhance availability, offer cross-market compliance and rapid equipment installation is to provide the customer a total solution with integrated backup designed into the end-user equipment.

The predominant change in the computer industry that makes this concept viable is the use of a distributed powering architecture for next generation systems. Although a 12 VDC bus is the suggested alternative for some low-end servers, the trend of developers is to concentrate around 48 V distribution schemes. Forty-eight volts is preferred because of lower distribution losses while maintaining guidelines for safety compliance. Driving forces for the distributed arrangements are higher power consumption, lower component voltages and higher distributed currents. Computing systems are becoming mirror images of telecom networking products and thus practical for the suggested improvements.

Before further defining this new concept, it may be beneficial to review present deployment strategies, their benefits and limitations.

Traditional data center A typical data center's components consist of separate power and equipment rooms partitioned by distribution breaker cabinets. Figure 1 shows the flow diagram of the power room.

A UPS provides backup. Static transfer switches may select either utility mains or backup generators as inputs. Backup generators traditionally are used to extend the reserve time of the internal batteries.

UPSs contain two conversion stages: The first rectifies the commercial AC mains, thus charging the reserve batteries; the second inverts battery DC to AC for office use. Should the UPS fail, an automatic transfer switch bypasses the conversion stages. For the duration of the fault, the equipment loses backup protection.

However, UPSs sometimes suffer from poor reliability and availability: Single points of failure disable the UPS. During UPS failure, backup is not available. Long cascaded strings of batteries reduce reliability. Poor efficiency also can be a problem. The UPS continuously burns about 20% of the energy consumed by the load.

Moreover, maintenance costs often are high. UPSs require about 10 times as many cascaded batteries as a 48 V system. Start-up and replacement costs can be expensive. Each office needs to be individually designed. Equipment capacity must be positioned in place for potential expansion needs. Where dual power feeds are used to improve the dependence on the utility power station, each feeder would require its own UPS.

But on the upside of UPSs, downstream equipment is not burdened by backup requirements.

CO battery plant A telecom CO's appearance resembles that of the data center. It has a battery plant that is contained in a divided power room. Distribution fuse panels separate the power plant from the switching and transmission equipment. But the similarity ends there.

The CO power plant takes advantage of direct use of the -48 VDC source by the feeder circuits. In one conversion stage, utility power is rectified directly to the -48 VDC level used by the office (Figure 2). The rectifiers are redundantly configured. The spare rectifier performs the dual functions of redundancy and battery charging. Batteries are connected directly to the -48 VDC bus. The available reserve time of the office is typically four to eight hours. Backup batteries and generators may be combined to provide this long-term hold-up.

The battery plant provides performance improvements over the UPS, including efficiency, reliability and availability. About a 20% efficiency improvement is derived from the elimination of the UPS's internal inversion stage and the need for an AC/DC supply in the post-equipment. Reliability improvements stem from the elimination of two conversion stages and the use of a 48 V battery string that reduces the number of batteries tied in series by an order of magnitude. Availability is improved by about a factor of 20, according to John Akerlund in an article for Intelec titled "-48 VDC Computer Equipment Topology - and Emerging Technology - Telia Network Surveys." Major contributors are the reduction of conversion stages, the N+1 rectification structure and the use of a substantially smaller number of batteries.

But there are drawbacks, such as the expense of distributing very high currents throughout the office. Moreover, the centralized approach requires numerous hierarchies of protection circuits. And each office needs to be custom-designed with sufficient capacity for future expansion.

Nonetheless, the option is efficient as a single conversion stage generates the office distribution bus. In addition, robustness and high availability are created by the N+1 rectifier plant and reductions in cascaded battery strings. And lower maintenance costs are mainly influenced by fewer battery strings and conversion stage reductions.

Both the data center and the CO require significant upfront capital investment to engineer each office and determine future expansion requirements; to provide the infrastructure of equipment, floor space and connectivity; and to back up two power feeds for the equipment, with each backup capable of powering the entire office.

Looking to the future From data presented in the same Intelec paper, it is a foregone conclusion that UPSs cannot meet the availability standards achieved by the telecom infrastructure. Computer manufacturers must consider improved solutions from those using UPSs to meet the availability objectives for next generation designs - solutions that would fit into the existing data center environment without a significant cost penalty.

One way to improve availability is to bring the backup circuitry either into the equipment or to partition the hardware so that a separate front-end power system with backup attaches to the equipment. The advantage of the latter partitioned structure is that computing and telecom equipment manufacturers pass the plant design on to the power vendor with established expertise in this area. Figure 3 shows a conceptual arrangement of such a product.

Localized conversion is similar in concept to the telecom office but with some notable exceptions. First, a DC generator would connect directly to the distribution bus. Positioning the generator across the distribution bus would provide the non-switched transfer sought by the Uptime Institute.

Second, the arrangement would eliminate the complex and costly bus-bar distribution and multiple protection arrangement of the present office.

There are even more far-reaching benefits. The localized reserve arrangement does not depend on either utility grid. It could be offered maintenance free or with hot-pluggable modular batteries. The complex, dual-feed duplicated battery plant arrangement could be retired for next generation offices by application of the distributed proposal.

Additional benefits can be realized as well. Backup is optimized to the performance requirements of the application. The configuration provides higher availability than traditional telecom. Adaptation into present offices is at the lowest cost to meet improvements in availability objectives. Note that with two backed-up utility feeds, this configuration may not provide substantially better availability. It is, however, a great choice as an add-on to an office that exhausted its available capacity.

The cost is lower and turnkey features are available for new and remote office installations. If deployed with a telecom-compliant distribution bus, the end-user equipment could be interfaced into the CO environment at the lowest cost and highest availability rate. Moreover, fault, including backup, is isolated within the equipment.

But a drawback is the potential proliferation of battery sizes as manufacturers optimize individual solutions.

A modular proposal The arrangement of Figure 3 suggests three packaging schemes with a three rectifier capacity shelf. The choice of three rectifiers within each shelf accommodates either single- or three-phase power feeders.

Three unique packaging arrangements provide cost-effective solutions for entry level single feed, intermediate level dual feed or high-availability dual feed with internal backup offerings. The equipment would be self-certified and simply bonded to the end-user system. Electrical interface would be limited to the DC bus cable and the control/communications between the two hardware platforms.

Although the concept could be deployed using any architecture, it does have some advantages when mated to distributed systems, especially to telecom -48 V.

Table 1 shows the cost tradeoffs between internal and external backup at the equipment level and for a small office. It assumes that a typical machine would consume about 10 kW. Table 1 also compares the cost difference for backup of a small office that could handle up to 10 machines. The comparison did not take into consideration the cost of distribution and centralized protection, site preparation, design planning, utilized floor space or operational savings.

The message being conveyed is that it takes extensive resources to plan, forecast and configure an office with centralized power plants. It also takes a significant investment to purchase the plant. All these costs can be avoided by deploying the distributed plant recommendation in new installations. And it can be done in a fraction of the time than what would be required to put together a new office. The distributed plant also is advantageous as an add-on in present data centers that have exceeded their capacity.

High-availability objectives and cross-market compliance can be met cost-effectively and more efficiently by using a distributed backup architecture designed in harmony with the equipment bay. By choosing a telecom-like -48V bus and offering a telecom-compliant product, this arrangement provides a cost-effective approach for placement of the product into COs.

The distributed plant architecture is similar in concept to the present CO structure. It has demonstrated reliability and availability unsurpassed by any other method. Further, it even exceeds the performance and availability of a CO because it does not rely on long distribution feeders and an extensive protection scheme.

Higher performance, availability improvements and cross-market compliance influence power system architectures for next generation systems.

Precise regulation and dynamic response are but two features driving performance improvements. Regulation gets much tighter as component voltage requirements are rapidly declining to around 0.8 to 3.3V. As supply voltages decrease, current consumption is steadily increasing in inverse proportion to these voltage reductions. High current consumption and fast load transition rates (di/dt) are forcing proximity of the converter near the powered load and thus dictate the use of distributed architectures.

Integrated voice and high-speed data services drive market convergence and the need for cross-market compliance of equipment designs. The issues are technology-independent. Designs will need to comply with both central office and data center requirements. Narrow market windows will preclude designing for one market and then redesigning to capture further market share.

Integrated services also will require uptime levels to be at least comparable with traditional telecom, that is, annual downtimes of less than 15 seconds. Drivers are crucial services such as 911 and demands for seamless service.

Time-to-market, product costs and expenditure reductions also are crucial market drivers. Technology advances rapidly make products obsolete. In this market, the product needs to be designed and placed into production fast and, simultaneously, meet instantaneous consumer demand.

The chosen powering architecture can play a crucial role in meeting these goals of performance, speed, cost and high availability.