Data Centre Energy Consumption Optimisation

[pinhodecarlos]

Hi all,

I'm an energy consultant and my next energy survey will be to an office which has a data centre.
Anyone here is an energy manager of this type of facility so we can have a discussion on the subject?
I've found a lot of guides but better talk with someone inside.

Thank you in advance,

Carlos

[pinhodecarlos]

Just for the matter I won't ask them to run Prime95...lol

My goal is not to calculate PUE but look at the physical infrastructure like cooling systems , lighting, schedules, pumps, etc as well as the data centre layout (position of the rack rows, extract and supply systems,). Also I would like to get feedback on heat recovery systems from the data centre, usage of free cooling.

[chalsall]

Originally Posted by pinhodecarlos

My goal is not to calculate PUE but look at the physical infrastructure like cooling systems , lighting, schedules, pumps, etc as well as the data centre layout (position of the rack rows, extract and supply systems,). Also I would like to get feedback on heat recovery systems from the data centre, usage of free cooling.

I wouldn't consider myself an expert, but I have built a few machine rooms.

One of the most important things to look at is the "hot isle / cold isle" concept, along with containment. What this basically comes down to is the AC'ed (cooled) air is fed into the front of a row of racks within an enclosed space, and the back (warm) air is extracted from a separate enclosed space.

The warm air can either be vented into an office space during the winter, or vented outside during the summer.

If you have multiple rows of racks, each pair of racks should be back-to-back; the "hot isle" being shared.

The containment space can be optimized such that the "hot isle" is contained and exhausted using fans (note the plural, for redundancy). The machine room itself is open (cold). The containment can be as crude as plastic sheeting, or very expensive metal plates.

A common mistake made is to not install "blanking plates" in racks where kit has not been installed. This results in a "short circuit" of the hot / cold air, which results in less than optimal energy usage.

Lastly, strongly suggest as much instrumentation as possible. All modern kit have temperature and power sensors which can be read remotely, and very inexpensive network enabled devices can be purchased with can report on power consumption, temperature and/or humidity. Personally I tend to use a dedicated 1U machine running Nagios and Cacti to constantly monitor each device.

I hope that helps.

P.S. I have never done water cooling, but others around here have.

P.P.S. I have done -48V DC power. Let me know if you need any advise on this.

[pinhodecarlos]

Thank you chalsall.

A few questions:

1) On heat recovery, warm air: is it direct ventilation inside space or do you add a filter on the pipe line? What's the usual air temperature? (just thinking about organic rankine cycles to produce electricity but I suppose the air temperature is below 50 degC). The analogy is the same applied on heat recovery from air compressors which can be used to supply hot stream air to warm spaces.

2) Do you meter your energy supply to the "IT load" before or after the PDU? If so do you have raw data so I can analyse?

3) Have you ever considered a CHP system?

4) Who from here have applied chiller to cool a data centre?

Carlos

[chalsall]

Originally Posted by pinhodecarlos

A few questions:

1. Six of the nine machine rooms I built were here in Barbados, where heating of the office spaces was not needed nor desired.

1.2. Filtering the warm air extracted from the machine room to introduce to the human's wouldn't be a bad idea; filtering the outside air entering the machine room would definitely be a good idea.

2. I meter everything I can. I'm not currently running a machine room, but attached is an image from Cacti running for a client a few years ago which showed that Barbados Light and Power were not providing the guaranteed voltage to the client. This data feed was from a Dell UPS (via SNMP) which showed that although the UPS was not being given the voltage promised, it was providing the steady voltage needed by the machines.

3. No, I have never considered a CHP system. Mostly because here in Barbados there is little opportunity for such systems to be cost effective.

Edit: Just so you know, the deep dips you see on the graph are right around noon during the work day (but not as much during the weekend). This was when everyone on the segment's AC units were working the hardest. Ironically BL&P fought hard against photovoltaics.

[chalsall]

Originally Posted by pinhodecarlos

just thinking about organic rankine cycles to produce electricity but I suppose the air temperature is below 50 degC)

I resonate with your thinking. I have often spent a great deal of time thinking about absorption cooling. Heat goes in, heat goes out.

Lithium Bromide. Hydrogen and Ammonia and H2O.

Einstein thought about this.

[GP2]

Google apparently got some good results by applying machine learning to optimize their data centers. I think it's still fairly experimental, but a very promising approach.

[chalsall]

Originally Posted by GP2

Google apparently got some good results by applying machine learning to optimize their data centers. I think it's still fairly experimental, but a very promising approach.

Please see https://static.googleusercontent.com...hive/42542.pdf

[chalsall]

Originally Posted by chalsall

Lithium Bromide. Hydrogen and Ammonia and H2O.

Please forgive me for this, but absorptive air conditioning might make sense.

Lithium Bromide reacts with metals and is easily contaminated. Ammonia is toxic to humans.

The thermal dynamic reaction is still worth considering. Many mobile homes have a cooler supplied by a heat source.

Please see this as an example.

[kladner]

OK. That has a steam feed for the heat source. It could also be a gas burner.

But what if you are trying to use solar water heating as a source? That would, at least, involve an added circulating pump. Could the system work effectively with a much smaller temperature gradient?

In a larger installation, the problems with ammonia in A/C use could be avoided by having a detached ammonia-water plant, which supplies cold water for the building.

Any ideas about calories removed kilowatt-hour compared to a standard compressor-driven refrigeration plant?

EDIT: I remember "gas" refrigerators which seemed to have something like a gas pilot light for a heat source. I don't know what they used for refrigerant and absorber.

[chalsall]

Originally Posted by kladner

But what if you are trying to use solar water heating as a source? That would, at least, involve an added circulating pump. Could the system work effectively with a much smaller temperature gradient?

Interestingly, flat-panel water solar collectors get only slightly less than the needed gradient under the best of conditions. You need large temperature gradients to do this kind of work.

Some interesting work has been done with what is known as "bubble pumps".

One thing I have spent some time thinking about is photovoltaic panels driving small compressors of the warm gas which are then able to heat the fluid solution to evaporate the molecules which can then be cooled and condensed.