Are we piggy-backing US data centre standards?


    If we have our own European data centre design guides and standards, should we be using them more? Ian Bitterlin provides a valuable insight into the standards, design guides and best practice currently available.

    shutterstock_463664303Are we Europeans piggy-backing North American data centre ‘Standards’? Let us consider this question for a moment. It is one I heard just the other day and the best answer is probably ‘yes’, with a small but growing ‘no’ caveat. At the root of the problem is widespread confusion as to what constitutes a ‘Standard’ since there is no authoritative definition.

    I personally differentiate between a ‘Standard’ (upper-case S) and a design guide or best practice document by defining it as a national or international document agreed by widespread industry consent and, hopefully, being as non-commercial based as is possible. The last point is subjective, as standards’ bodies around the world depend upon expertise to draft their documents and that expertise nearly always must come from industry, and so the technical input always has a commercial agenda.

    I am reliably informed that anything can be considered a standard if it is accepted by manufacturers and customers alike, and it does not even have to be written down formally. However for the purposes of this article, we shall take the slighter higher moral ground and take standard to mean that issued under the ISO/IEC/ANSI/EN/BS banners.

    So, the first thing to clear up is that all the plant and materials used to construct data centre infrastructure are covered by individual standards and industry certifications – such as IEC for UPS performance or Eurovent certification for cooling plant such as CRACs and chillers etc. The standards that we are most interested in for data centre design and operation are the design guides and best practices such as EN50600 and the EU Code of Conduct, so let’s look at what is available.

    Our designs are based upon three primary considerations, all of which originate in the US:

    • Power quality requirements for ICT hardware

    • Thermal requirements for the inlet of same

    • Availability of ICT service through redundancy and concurrent maintainability

    The first, voltage fidelity, is only published in one place, the CEBMA/ITIC Curve in the ANSI Standard IEEE-1100 and 466, published in the Color Series ‘Emerald’ book. It is, by universal agreement, now outdated and being investigated by The Green Grid and EPRI in the US. The ‘latest’ version is 1997, which shows a 20ms zero-voltage tolerance but, in the pursuit of energy efficient power supplies, is now much shorter and often regarded as dependent upon ICT load and, perhaps, influenced by the choice of single or dual corded hardware.

    So, in the absence of anything else we are solely dependent upon The Green Grid/EPRI influencing an update of the ANSI IEEE standard.

    It is highly unlikely that any European standards or industry body would try to supplant this power quality requirement but there is a voltage tolerance curve in the European static-UPS standard, IEC 62040-3, that bears no resemblance whatsoever to the actual hardware that is now designed in the US and manufactured in the Far East, and probably never did. The increase in power supply efficiency (Energy Star, 90+ etc) has been dramatic in recent years although users must pay extra and many don’t.

    The second consideration, thermal guidelines for temperature and humidity, is, again, only published in one place, the Thermal Guidelines of ASHRAE. This is issued and regularly revised by TC9.9, which constituted with a key membership of ICT hardware OEMs. Again, no European body would be likely to try to supplant the guidelines, not least because the drive for lower energy cooling has been rapid. Not all users are prepared to be early adopters and even the ‘allowable’ range of temperature and humidity in the 2011 version is still regarded by many users as avant-garde.

    The third consideration, designing for availability with/without concurrent maintainability and fault tolerance, now has many protagonists but was started by the Uptime Institute, in the US. Still regarded by many, especially in the developing data centre markets, its ‘Tier Classification’ in four levels is often referred to even though UTI certification is rare (compared with the installed base) in Western and Northern Europe.

    Worthy beginnings

    The Uptime Institute had noteworthy beginnings: The founders worked with IBM to ‘innovate’ the dual-cord load and then went one step further by publishing a series of white papers that described a classification system that suggested how to organise power and cooling to achieve concurrent maintainable (Tier III) and fault tolerant (Tier IV) infrastructure.

    Having started the ball rolling, it has gradually refined its requirements and, although logical, it has some very particular features required from the emergency standby generators that other classification systems have not generally followed. TUI classification concentrates on power and cooling. The point is that it is a ‘private’ scheme, without peer review other than Uptime clients, that cannot be accredited or certified by anyone else. It is certainly a design guide, not a ‘Standard’, and it is not ‘European’ in the sense of our original question. It is generally lacking in detail (as opposed to the concepts) so it does travel well across continents and regulatory jurisdictions, perhaps except for the neutral conductor in European electrical systems. TUI now goes further into certification of ‘operations’ – where most data centre failures are rooted.

    Following on from TUI came the ANSI Standard issued by the ‘International’ Telecom Association, TIA-942, very much a US-based design guide. For many years, this more proscriptive and wordy document was based entirely on the four Tiers of TUI but more recently they fell out (allegedly over third-party certification) and TIA942-A was reissued with a change in wording from ‘Tier I-IV’ to ‘Data Centre Type I-IV’, and removal of ‘with grateful thanks to TUI’ in the preface.

    Because it is more proscriptive, less can be applied outside of North America but anyone could purchase a copy and, given the right qualifications, expertise, experience and PI insurance and, if acceptable to the client, they could audit a facility and issue a ‘Conformity to TIA-942’ statement – albeit under their own name, not TIA.

    BICSI also contains an infrastructure design guide, 001, and is also issued as an ANSI Standard. With its roots in ICT cabling infrastructure, this is the least used in Europe for power and cooling but it is also based on four Levels, 1-4, once you ignore Level 0, which is a data centre without UPS or generator, so not being much of a facility in the first place.

    Finally, for Europe, we have EN50600. This is largely now published in multiple parts and is a design guide covering the whole facility, not just power and cooling. It is wholly European but is not yet fully distributed nor used. It is expensive to buy but more importantly it has no established track record nor pedigree. It is nonproscriptive and some say this is a weakness.

    Again, anyone can purchase a copy and audit a facility and issue a ‘conformity’ statement under their own name, if the client values that. One point worth noting (and I don’t know yet if this is a strength or weakness) is that it has four levels, 1-4, of ‘Availability Class’ in each of power, cooling and connectivity etc but does not apply the ‘lowest common denominator’ to the overall facility Class as does TUI.

    It also had in its first issue a problematic Cooling part in that some idiot convinced the editor that Class 4 dual-bus cooling was not possible so the original document had five levels, 1-4 plus ‘enhanced 4’. This is to be corrected in the future editions. How it came to be published like that is still a mystery but it will all be alright in the end. There is one silly little problem that can’t be corrected and that is the target thermal guidelines should be included but the EN system won’t reference a source like ASHRAE as it is, at worst, a ‘foreign’ trade association and, at best, a document that is regularly updated.

    So, we DO have a European design guide all of our own making and we should use it more. The advantage for us in the UK is that it is first written and published in English, so it reads well.

    Common features

    Have you spotted the common feature between Uptime, TIA and BICSI? The common feature is Four Levels/Types/Classes and the answer is simple – there are only four possible architectures in a power system that feed a dual cord load:

    1. Single path without redundancy so that failure of any component or path or maintenance shut-down requires a load shut-down

    2. Single path with redundant components making some maintenance concurrent

    3. Dual path, one active path with redundant components plus one wrap-around passive path for fast recovery

    4. Dual path, both active (with or without redundant components) with separate paths and fire-cells etc

    Lastly, we do have an international ISO/IEC that the UK has an input into, for ‘Resource Effective Data Centres’, namely ISO/IEC 30134 in multi-parts. This will adopt (sometimes with changes/edits) the data centre metrics written by industry association, particularly from The Green Grid, such as PUE (now ISO 30134-2), and water, carbon, renewable energy, reuse of waste heat etc. Even the EU CoC (not in any way a design guide) and TGG Data Centre Maturity Model are being adopted in the wider ‘Standards’ world as technical reports etc. As long as all this work remains non-commercial we shall be well served


    Please enter your comment!
    Please enter your name here