Ian Bitterlin's Blog http://www.datacenterdynamics.com/ME2/Console/XmlSyndication/Display/RSS.asp?xsid=2498E6394B1D418BA8C3834418554DC3 Ian Bitterlin's blog IT load on bypass to mains or genset? http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=32D4409E4DDE4C2EBF8025E76A93271B I had a question regarding what is ‘better’ (or even ‘accepted practice’) between genset and grid when a UPS is in bypass and if it was documented as ‘preferred’ practice anywhere. I could confirm that ‘genset’ is the accepted normal practice but I couldn’t say if or where that is written down!  Does anyone know any better? What I can give are the good reasons why genset power is preferable: The main risks with genset reliability are ‘starting’ and ‘paralleling’.  Once these events are successfully accomplished there only remains (for a few hours of running) the matter of fuel – supply (quantity & pump arrangements) and quality. Once the set(s) is/are up-and-running and the fuel in the day-tank is guaranteed, almost nothing can stop it for the duration for at least half of the day-tank capacity.  Its generally ‘half’ if the fuel return feed from the injector rack is re-circulated to the day-tank – you don’t want the day-tank oil getting too hot The likely MTBF of genset power (once running) is around 5,000h On the other hand there is one simple reason why grid-power is not recommended: You have no control over its ‘continuity’ You have no predictive methods to cover its ‘quality’ and you expose the load to the increased risk of transient surges (overvoltage spikes generated on the grid) It will, on average, deviate outside of the CBEMA ITC curve, with an MTBF of 250h (20x times more likely to fail than a genset supply) There is one area to be careful of – loads with leading power factors.  IF the load is capacitive (leading) then be careful not to load the genset above the limit of its export-kVAR capacity.  This will usually be around 25-30% load if the load is 0.9leading. Below I have cut-and-paste from my DCD Blog of last year on this subject Gensets are heading into stormy waters With the latest generation of power factor corrected switched-mode power supplies comes low load current harmonics (typically 8-10% THID) but a very unhelpful leading power factor of around 0.95.  The leading power factor comes about from partial load on redundant dual-cord server connections.  Gone are the heady Triplen days of 70% THID and PF=0.6. As a result marketing men have swung into action and made their claims for leading power factor capabilities for various flavours of UPS system.  That can be the topic for another day. What is not generally discussed (yet) is the case for the average emergency generator set with its 15% impedance alternator and inability to generate large quantities of kVAr.  Absorbing kVAr (to 0.8 lagging PF and beyond) has never been an issue but generating kVAr is another matter altogether.  In fact most standard generators simply can’t do it and regulate the output voltage at the same time. This is made worse by the application of VSDs to the mechanical load – no longer is there a chunk of 0.8PF motor load to throw onto your genset before you apply the IT load.  That is now harmonic rich and near unity power factor.  Oh happy day … And?  I hear you ask.  Well, soon gone will be the days when you will be able to run your load on generator whilst carrying out essential maintenance on your UPS output distribution system.  And as for a trip to bypass when on genset – oh no.  It will not be an operational scenario to ‘ever’ supply the load with raw genset power – the voltage will be all over the place and all of your protection relays will race each other to trip and drop the load. The solution?  Pay more for a higher spec alternator (at least a larger frame size) and genset electrical control system.   Tue, 17 Aug 2010 10:24:42 EST 32D4409E4DDE4C2EBF8025E76A93271B I had a question regarding what is ‘better’ (or even ‘accepted practice’) between genset and grid when a UPS is in bypass and if it was documented as ‘preferred’ practice anywhere.

I could confirm that ‘genset’ is the accepted normal practice but I couldn’t say if or where that is written down!  Does anyone know any better?

What I can give are the good reasons why genset power is preferable:

  • The main risks with genset reliability are ‘starting’ and ‘paralleling’.  Once these events are successfully accomplished there only remains (for a few hours of running) the matter of fuel – supply (quantity & pump arrangements) and quality.
  • Once the set(s) is/are up-and-running and the fuel in the day-tank is guaranteed, almost nothing can stop it for the duration for at least half of the day-tank capacity.  Its generally ‘half’ if the fuel return feed from the injector rack is re-circulated to the day-tank – you don’t want the day-tank oil getting too hot
  • The likely MTBF of genset power (once running) is around 5,000h

On the other hand there is one simple reason why grid-power is not recommended:

  • You have no control over its ‘continuity’
  • You have no predictive methods to cover its ‘quality’ and you expose the load to the increased risk of transient surges (overvoltage spikes generated on the grid)
  • It will, on average, deviate outside of the CBEMA ITC curve, with an MTBF of 250h (20x times more likely to fail than a genset supply)

There is one area to be careful of – loads with leading power factors.  IF the load is capacitive (leading) then be careful not to load the genset above the limit of its export-kVAR capacity.  This will usually be around 25-30% load if the load is 0.9leading.

Below I have cut-and-paste from my DCD Blog of last year on this subject

Gensets are heading into stormy waters

With the latest generation of power factor corrected switched-mode power supplies comes low load current harmonics (typically 8-10% THID) but a very unhelpful leading power factor of around 0.95.  The leading power factor comes about from partial load on redundant dual-cord server connections.  Gone are the heady Triplen days of 70% THID and PF=0.6.

As a result marketing men have swung into action and made their claims for leading power factor capabilities for various flavours of UPS system.  That can be the topic for another day.

What is not generally discussed (yet) is the case for the average emergency generator set with its 15% impedance alternator and inability to generate large quantities of kVAr.  Absorbing kVAr (to 0.8 lagging PF and beyond) has never been an issue but generating kVAr is another matter altogether.  In fact most standard generators simply can’t do it and regulate the output voltage at the same time.

This is made worse by the application of VSDs to the mechanical load – no longer is there a chunk of 0.8PF motor load to throw onto your genset before you apply the IT load.  That is now harmonic rich and near unity power factor.  Oh happy day …

And?  I hear you ask.  Well, soon gone will be the days when you will be able to run your load on generator whilst carrying out essential maintenance on your UPS output distribution system.  And as for a trip to bypass when on genset – oh no.  It will not be an operational scenario to ‘ever’ supply the load with raw genset power – the voltage will be all over the place and all of your protection relays will race each other to trip and drop the load.

The solution?  Pay more for a higher spec alternator (at least a larger frame size) and genset electrical control system.

 

]]> Servers in a tent? Not so off-the-wall now? … what wall? http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=1D3872E4EE654DCA8C380202289D5798 The famous ‘servers in a tent’ trial - with a hardly discernable increase in IT hardware failure other than autumn leaves blocking the fans - was not long ago and caused a great deal of discussion and some considerable consternation. I guess the candles burned bright in many a precision air-conditioning OEM’s marketing department those nights! Just a couple of years on it seems incredible that fresh-air cooling is almost becoming ‘standard’ – and certainly not ‘wacky’. The drive for energy efficiency (saving money not being ‘green’) has fuelled the market seed-change. OK, there is still the vast majority of new projects using conventional CRACs and out-of-date set-points but the genie is definitely out of the bottle. Is it only a matter of time…? Wed, 11 Aug 2010 06:57:08 EST 1D3872E4EE654DCA8C380202289D5798 Hottest day? Gentleman, start your engines! http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=F105A6D1E65C4AFFB640A7ABFDF8CB20 At long last we are working on a project where the emergency generator sets are sized larger than the grid supply so as to maximise the grid utilisation for IT load and yet still feed the maximum cooling load on the hottest days of the year. If you utilise the fullest range of server inlet temperature set-points allowable under the latest ASHREA recommendations the opportunities for free-cooling (compressor free) are enhanced. With the best-in-class free-cooling chillers in the UK climate the compressor hours per year can be in the 150-200h range and if fresh-air cooling is applied this can fall to 50-100h. As the compressor energy may be in the order of 15-20% of the load to be cooled the reservation of so much grid capacity for only 2-3% of the year is patently wasteful of a valuable resource. Hence the rather anti-intuitive strategy of starting the generators when the peak load occurs. Despite burning fuel oil I have calculated that the overall carbon footprint is lower – not to mention the better utilisation of the carbon invested in the gensets themselves. Lovely. Wed, 11 Aug 2010 06:52:14 EST F105A6D1E65C4AFFB640A7ABFDF8CB20 Here we go again - solar powered data-centre hype http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=FB2D3DC414624C21BDE3F291D1E9DD5C At least the press release below didnt try to hide anything under a 'green' cload of smoke but who can define a data-centre (even a 'micro' one) as needing 1.84kW of power? I suppose that we should be grateful that someone can multiply 8x230W and reach 1840W. This must be a 'nano-centre' - surely a new catagory in the awards programme? Here is the press release: Solar-powered data centre launches in Green Internet Network A rooftop solar-powered data centre has been connected to Canada’s first “green” powered internet network. On June 29, 2010, Cybera, with national partners CANARIE and the GreenStar Network (GSN) Project, connected the Calgary node, which is managed by Cybera, will draw more than 1,840 watts of power from eight solar panels (230 watts each) installed on roof space donated by Calgary Technologies in the Alastair Ross Technology Centre. Over the next few months, the GSN Project will connect five different nodes across Canada, each powered by renewable energy sources as they store and transfer research data for pilot user groups. The GreenStar nodes will act as small data centres and will draw power from renewable energy sources such as wind, water and the sun. If the node uses up its power supply before it is able to recharge, the data will be seamlessly transferred along the network to another operational node. Source: Scientific Computing Tue, 03 Aug 2010 05:11:55 EST FB2D3DC414624C21BDE3F291D1E9DD5C Old hat but very few takers? http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=568EF8BEFFB34DCDB79BC46059CDDFF7 Recently I spoke to an industry guru, Bob Sullivan (Dr Bob), and commented that banging on about aisle containment, blanking plates, hole stopping and raised temperatures etc was starting to sound jaded. He made a very interesting reply; 'maybe, but until people actually start doing it we have to keep on giving the message'. He is right - I visit a large number of facilities and I still hardly ever see the 'best' practice. Why is that? It cant be cost because the entry cost is low and the return on investment is very rapid, often <6 months. Mon, 19 Jul 2010 04:23:16 EST 568EF8BEFFB34DCDB79BC46059CDDFF7 I see that HP Labs are talking bullshit? http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=C6A4B1E127D244838D2C803898459304 HP Labs have announced that 10,000 cows can produce enough manure to generate power for a 1MW data-centre. I'm sure they are right but I would prefer them to work on the problem that their IT server boxes consume 60% of their rated power when doing exactly no useful computing work. No doubt they are trying to raise a smoke screen - best achieved by burning the manure? Thu, 01 Jul 2010 08:05:52 EST C6A4B1E127D244838D2C803898459304 Name plate data mystery solved at last? http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=6EACC4BC03234245BA0E23F9F89F075D We have all read how the average chip utilisation is 5% and, for more than a decade, we have all discussed the huge discrepancy in name-plate VS actual power draw of Servers. At the same time the Server OEMs have been introducing high-efficiency versions – with a combination of low loss power conversion and low wattage chips. The power converters have taken flack over the years on three fronts – high harmonics, low efficiency (especially at partial load) and leading power factors. Most people I know (including me!) use a rule-of-thumb that says IT boxes consume 1/3rd to 2/3rd of their name-plate ratings. Hence a 7U blade chassis that says it wants 5.5kW often draws only 2-2.5kW when installed. The ‘problem’ for the IT OEMs on the name-plate issue – especially so on blade chassis topology - is that they sell a huge number of options and the name-plate power draw covers fully configured machines. This is aggravated by the wide frequency and voltage tolerances – where you can often see the input range to be 45-65Hz and 208-240V. I’ve even spotted 110-250V but didn’t believe it at the time. However if we step back and consider where the power goes inside the box we can reach a figure that supports the ‘rule-of-thumb’. The power goes via the power supply conversion (still often only 65-70% efficient at low load) to the processor chip(s), the memory (chips and spinning hard drives), the I/O and the on-board cooling fans. The result is that when you turn the average server on it draws 2/3rd of its power with the processor doing no useful work at all. As you will hardly ever reach 10% IT load the power draw remains constant – spinning the drives and the fans and powering the I/O. In fact a server does a lot of things within that 5% that are not actually ‘real’ processing – virus scans and indexing for example. As the chip hardly every climbs above 5% utilisation the power stays constant (at 2/3rd). So the real target for energy saving is not PUE but the server box itself. Power management, lower power drives (solid-state?) and I/O seem to be the key. If you want a longer version of this I have written a white paper entitled ‘PUE is not the holy grail’ that you should be able to download soon from www.prismpower.co.uk – enjoy. Fri, 04 Jun 2010 11:24:45 EST 6EACC4BC03234245BA0E23F9F89F075D Indoor Clouds? Rains inside the critical space? http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=3FE40DBDFF234CB6834357509C72EAA7 Is it just me or have we all witnessed a gradual change in the definition of what ‘cloud computing’ actually IS? Clearly ‘cloud’ is still a virulent buzz-word in the IT industry and has been morphed into such wonders of the modern world as G-Cloud with the ‘G’ standing for Government – or is that Laurel & Hardy? Anyway, when heralded as the latest thing since Wonderloaf, the concept of ‘cloud’ was that your data and processes would be ‘out there’ in some virtual nebulous network on the internet. Indeed it would move, float, evaporate and condense at the drop of a weather forecast. Your virtual server would reside ‘somewhere’ and the overall internet load would be automatically managed and optimised. Great theories abounded about enhanced resilience and the death of the high-performance data-centre – whilst horror story writers pedalled the view that data-security would be impossible. On the other hand pundits (tits who pun?) claimed that we would have an outbreak of thousands of small low-resilience Tier I data-rooms and that security could actually be higher because a targeted hacker would not know ‘where’ to look for your precious bytes. But if you look around and read the press releases of 2010 you actually come across very many ‘cloud enabled’ services being touted by IT service providers who run, or are hosted in, well defined physical infrastructures, i.e. data centres. Surely (and, yes, don’t call me Shirley) if you know where your data &/or process (and the server that it runs on) actually ‘is’ then that is not ‘Cloud’ but ‘Managed Hosting’? The Managed Hosting boys must be licking their lips at the sight of the dark, money laden, tumuli heading their way, and good luck to them, but at least it will stimulate IT services out-sourcing, which (by happy coincidence) should be what G-Cloud really ends up as. We, the general plebiscites, have been duped – the clouds have moved indoors. Or we can console ourselves with reading Wordsworth; wandering like a cloud on high … (with no apology for the misquote). Please note the resolve shown not to mention the song ‘raindrops keep falling on my head’ nor ‘somewhere, over the rainbow’ – I still could be Dorothy! Thu, 13 May 2010 10:30:16 EST 3FE40DBDFF234CB6834357509C72EAA7 Green is the colour of money… http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=CF32E6D0FA0A488D864DB2C25AB63AB2 The term ‘green computing’ means different things to different people: To some it is about not regularly refreshing hardware so as to avoid hazardous waste reprocessing by children in the third-world – an argument put forward by Greenpeace. To others refreshing hardware enables the user to take advantage of better power efficiency and pays for itself within two years and they argue that the waste recycling laws are for the government to police and enforce, e.g. the redundant products should not get as far as the third-world and should be correctly reprocessed at source. To a minority ‘green computing’ has become a software-hardware performance issue, i.e. does the application software use the processor power in the most efficient way. From the hardware point of view the computing performance (e.g. calculations per second) per Watt consumed is just starting to become a viable metric with ‘standardised’ software routines for testing Servers. To most corporate end-users ‘green IT’ is simply related to saving operational energy costs and good PR in the annual report – although if it saves energy the most cynical of motives are perfectly acceptable. Green is the colour of money. To M&E designers of data-centres ‘green’ has become embodied in the difference between the power consumed by the IT hardware and the power consumed by the whole facility that supports the computing function – the, so called, Power Usage Efficiency (PUE) metric. This metric has been adopted from The Green Grid by the EU Code-of-Conduct and will, no doubt, eventually be at the nub of the performance measurement for the CRC for data facilities. The PUE is the total incoming power divided by the IT power consumed, e.g. a facility that supports a 500kW IT load and draws 1000kW from the grid has a PUE of 2.0. The difference between a ‘good’ facility having a low PUE (e.g. 1.5) and a ‘bad’ facility with a high PUE (e.g. 3) is almost entirely a matter of the efficiency of the cooling system as long as it is fully loaded. So, you may think that the whole spectrum of energy sustainability in data-centres is covered by the present initiatives but, unfortunately, that is not strictly true. As its highest level in classic efficiency definition the data-centre consumes energy (work in) but turns all of the energy into heat (no work out) and is, therefore, zero efficient – so how can we apply the rules of sustainability to such a mechanism? The first rule of energy sustainability is to reduce consumption. However data-centre load power consumption, despite massive increases in processing capacity per Watt, continues to climb. The growth in on-line data-storage is one culprit but very few people ever challenge ‘what’ the data-centre actually ‘does’. It is easy to prove that large data-centres are far more ‘efficient’ (consume less power) than many small data-rooms embedded in office facilities and it is easy to prove that some data-centres enable low carbon solutions – internet grocery shopping for example has one delivery vehicle route compared to several tens of car journeys to the out-of-town hypermarket. At the bottom of the ‘useful’ list may be social web-sites but where would business be today without email and the internet? Without data-centres there would be no internet. The second rule of energy sustainability is to improve the efficiency of the process and reduce waste and you would think that improvement in PUE would cover this entirely – but you would be wrong. The measurement of PUE is based on the power consumed by the IT hardware (the ‘box’) compared to all the other consumers of power; cooling, ventilation, humidification, security, lighting, UPS losses and power distribution losses. The cooling system consumption (often between 30-40% of the total facility power consumption) is by far the largest target for improvement and is driven by the room temperature and ability of the system to utilise the external ambient temperature, so called ‘free cooling’. So M&E engineers are continually striving for customer acceptance of warmer rooms and a slightly higher investment in free-cooling plant. At present energy cost levels the short RoI is an overwhelming argument in favour of ‘green’ engineering solutions – and energy costs will only rise in the future. However there is one missing element in the chain; the IT hardware itself and its power consumption compared to its computing load. Inside the ‘box’ is a power supply system that takes the incoming voltage (normally 230V AC but can be high voltage DC) and transforms and regulates it right down to the c1VDC that the microprocessor uses. The average examples of that power supply system have increased in efficiency over the past years from 40% to 70% and the heat losses have to be rejected by fans that also consume power and add to the inefficiency. As we can see the IT ‘box’ should be getting the same pressure to improve as the M&E plant but that is rarely discussed and usually ignored (as a subject area of too much complexity and conflict?) by such instruments as the EU CoC. However, before we challenge the IT box power efficiency too much there is one further problem to face up to. The average IT Server is loaded (computing) to 60%. At these levels no one can expect the on-board power supply to run at peak efficiency so maybe we should start with increasing the machine utilisation? Tue, 04 May 2010 04:29:54 EST CF32E6D0FA0A488D864DB2C25AB63AB2 60%. At these levels no one can expect the on-board power supply to run at peak efficiency so maybe we should start with increasing the machine utilisation? ]]> EPA can’t interpret numbers? Energy Star program at risk from innumeracy. http://www.datacenterdynamics.com/ME2/Audiences/dirmod.asp?sid=&nm=&AudID=E5BD2FF22AF74DF3A0D5F4E519A61511&type=Blog&mod=View+Topic&mid=67D6564029914AD3B204AD35D8F5F780&tier=7&id=699996722B474149B5426DBBDC7AEE7E The headline read ‘A data center’s uptime has no impact on its energy efficiency’ and an EPA Energy Star spokesperson said that ‘the data didn’t show a significant relationship between Tier and energy use’. Gasp, shock horror. The little table published with this (piffle) was from the upcoming EPA attempt to rate data-centre efficiency on a 0-100 scale – but why PUE isn’t enough, don’t ask me. I hope they include water use in it. Anyway the table showed a total of 108 facilities split into four classes, Tier I-IV, and the average PUE for each Tier. The table itself was interesting as only 2 facilities owned up to being Tier I whist a whopping 21% claimed to be Tier IV. Anyway, the numbers in the table read 1.81, 1.92 and 1.96 for Tier II, III and IV respectively. There was no entry for Tier I. It also showed that TierI+II averaged 1.83 whilst Tier III+IV averaged 1.94. At GB£0.10/kWh the operating cost difference is £96,000pa per MW of IT load. That’s pretty significant to me. Mind you, how you average ‘2xTierI with no reported PUE’ with ‘27xTierII at 1.81 PUE’ and get 29 facilities with 1.83 is beyond my maths. I don’t know about you but, despite the strange arithmetic, those figures show a clear relationship between Tier & PUE. What is the EPA smoking these days or doesn’t the ability to count ‘count’ anymore? Tue, 13 Apr 2010 06:16:12 EST 699996722B474149B5426DBBDC7AEE7E