Data-centre Tier Classifications and their impact on UPS power

Standards

Developed by The Uptime Institute (TUI), the Tier Classification and Performance Standard provides a reference for objective comparison in such situations. TUI has for nearly 20 years sponsored research and practical studies into data centre design, operation and resultant resilience. The Standard centres on assessing a data centre’s availability; its ability to remain functional throughout concurrent maintenance or component failures. Tier Classification defines four levels of availability for a complete site, ranging from the basic Tier l to the highest-availability Tier lV.

A later addition to TUI is a data-centre ‘standard’ in ANSI/TIA-942-2005 Telecommunications Infrastructure Standard for Data Centers, issued by the Telecommunications Industry Association. This follows the same Tier I-IV format and draws heavily on The Uptime Institute publications but extends the detail, especially in connectivity, and is more proscriptive. It is entirely a USA-centric ANSI specification, but is a very useful guide beyond the reach of ANSI. Note that TIA-942 was specifically written for telecom related data-centre environments with power densities to 2.7kW/m2.

BICSI published another US-centric design guide which introduced a ‘fifth’ Tier. However this was Tier ‘0’ and described a data centre without UPS or generator support that most observers would not classify as a data-centre.

CENELEC is preparing a new European Standard, EN 50600, for data centre infrastructure which will also be based on four levels (Classes rather than Tiers) of Availability.

TUI Tier Classification System

The true significance of a data centre’s TUI Tier rating arises because it depends on all aspects of the site’s infrastructure that impact operation, including both subsystems installed and operational sustainability. This in turn involves many factors including topology decisions about robustness and operability, construction implementation, and site management and staffing. The rating is also constrained by the infrastructure’s lowest-rated aspect. For example, a site with a robust Tier lV UPS installation and a Tier ll chilled water system will be limited to a Tier ll site rating. Another point is that the Standard is not prescriptive. Each classification is based on performance-based outcome requirements, which can be satisfied by more than one design topology.

Table 1 shows the key parameters for the Standard’s four Tiers. The system is limited to four Tiers through being based on two variants each of two design techniques – dual cording and redundancy. A fifth Tier could theoretically be defined, for example by specifying triple-corded loads, but such approaches are not taken in practice.

Table 1: Tier requirements summary

Dual cording refers to IT equipment which has two independent power inputs, each sufficient to meet its entire power requirement. The founders of TUI developed this concept, which formed the basis of the tier classification system they went on to produce. If any event, anywhere on the data centre site – or possibly beyond it – should interrupt one power feed, the other can continue to drive the IT load, without the load suffering power interruption or loss. This is redundancy at the power distribution path level.

Redundancy brings the same type of resilience to failure at a system level, using a technique known as ‘N+1’. In a UPS system, for example, if a configuration with N modules is sufficient to entirely support a critical load, then a configuration containing N+1 modules would continue to support the load without power interruption or loss even if one module fails.

Accordingly, a Tier l basic data centre has non-redundant capacity components and a single, non-redundant power distribution path serving the ICT equipment. Within such a system, conditioned power to the load is lost whenever the UPS has to be shut down for maintenance, or if a fault in the UPS or power line occurs. Tier ll is an improvement on this, as it introduces redundancy in the infrastructure components as indicated by N+1 in Table 1. This gives protection from a single failure and allows a degree of concurrent maintenance.

Tier lll has a dual bus power system connected to critical loads with dual power inputs. It includes Tier ll type redundant capacity component configuration. Typically, only one power bus serves the ICT equipment at any one time. Concurrent maintenance is fully supported. Tier lV, the highest rating, is classified as a fault tolerant site infrastructure. It comprises two physically segregated active power paths, each with an independent Tier ll system. It offers very high availability, concurrent maintenance and near total fault tolerance.

Availability

The ultimate point of achieving a Tier classification is the Availability that it brings. TUI has specified Availability benchmarks for each Tier, which are shown in Table 2. TUI considers a data centre to have 16 systems, including UPSs, which are critical to its operation. For an installation to be rated to a given Tier, each of its critical systems must achieve that Tier’s Availability benchmark.

Table 2: Availability benchmarks for Tier Classification

Availability can be increased either by increasing Mean Time Between Failures (MTBF) or by decreasing Mean Time To Repair (MTTR). It can be calculated using the equation

However, because the data centre’s overall Availability is based on that of each of its critical subsystems, each subsystem must achieve a much higher performance, based on A% raised to the power of 16. In other words, every subsystem has to achieve 99.9994% - the magic Five-Nines – for the data centre to receive the ultimate Tier lV site rating.

The equation shows how MTBF and MTTR can both be varied to produce a target Availability. However in reality MTTR values must be kept to minimal levels to generate Availability levels sufficient for the higher Tiers. Additionally, true MTTR calculations must allow for travel time to site and depend on the availability of any required spare parts together with first-time, fast fix capability. For example, Tier lll classification cannot tolerate travel times of more than 4 hours to site to maintain the required Availability, even with MTBFs in the 200,000 – 400,000 h range. These figures highlight the need for a first-class service organisation with 24/7 remote monitoring, diagnostics and tele-assisted support via data-connectivity.

Increased Availability comes at a cost, in terms of both capital and operating expenditure, for Tier classified power systems. Table 3 shows the relative capital costs of different Tiers, using the Tier l figures as a base reference.

Tier A% Power System MTBF COST
l 99.98333% 1 1
ll 99.98547% 1 1.6
lll 99.99983% 45 2.2
lV 99.99999% 2450 3.0

Table 3: Relative capital costs and performance for different Tier levels

Tier classification affects operating costs as well, particularly in data centres using legacy power installations comprising large monolithic UPS systems. Tier lV dual-bus 2(N+1) architecture inevitably subjects them to partial loading, where they operate very inefficiently. This situation is exacerbated by the low load conditions normal to data centres, reducing the load per UPS module to possibly below 5%. This results in extremely low power system efficiency.

Impact of modern UPS technology

However several developments have mitigated this traditional downside of Tier lV. Its requirements were reduced by TUI from the double-redundant 2(N+1) to 2N, where each system is fully redundant for the other. This raised the UPS module load by several percentage points. Additionally, modern transformerless UPS technology has brought much greater efficiency at all loads, achieving over 96% even at 50% loading.

Transformerless technology, due to its size and weight saving, also allows UPSs to be configured incrementally and right sized to their load. This ensures high UPS loading and further efficiency. Additionally, modern systems offer Eco-mode options which enable efficiencies of over 98% even at 10% load. Even users with reservations about subjecting their load entirely to eco-mode may accept using it with one of their two power buses.

Overall, Tier lV classification brings many significant advantages to data centres. In addition to increasing statistical Availability by several magnitudes, it can also provide resilience to human error – a factor responsible for 60 – 70% of all data centre failures, according to most reports. It is the most resilient possible power architecture, by a factor of over 1000x. Modern UPS technology has removed its efficiency and operating cost disadvantages. It still has a high CapEx – typically a 35 – 40% premium over Tier lll – yet its ultra-high level of availability means critical data centre operators have little choice but to follow it, or in future its European alternative specified by EN 50600.