How do we measure the DIC?

The DIC was intro­duced in the previ­ous edition of Cloud­Works and Data­cen­ter­Works. The purpose of this measure­ment is to provide insight into how effi­ciently IT handles the ‘power’ in the data center.
While it is not clear at this time what the effi­ciency is, it is esti­mated that 60 to 80% of all energy to the data centers is consumed without real IT perfor­mance being deliv­ered. In short, doing nothing in IT costs a lot of energy.

The DIC has been defined by Dirk Harry­van and the under­signed as:
The energy consumed at idle, divided by the total energy consump­tion of the IT hardware

Idle is the state of the CPU that it is ‘waiting’ for exter­nal work. In idle state, the CPU is active for example for the oper­at­ing system and the like, but it does not perform any exter­nal work. Idle can be very short but also longer — from millisec­onds to long periods.

In figure 1 I explain the DIC calcu­la­tion. We assume that the distri­b­u­tion of power is for IT: servers 70%, storage 20% and 10% for the network. If the servers are active, that is useful and there­fore green. If the server is in idle state, then red. Red because nothing useful is running. Below that is the distri­b­u­tion between active and idle. In this example 50–50%. Again the equal color distri­b­u­tion. In this example, the IT hard­ware ‘uses’ in idle state just as much energy (70%) as in active state. The surface of power consump­tion and active/idle distri­b­u­tion is now: 50% of the power consump­tion is consumed in idle state and 50% in active state. The DIC is defined as a number always between 0 and 10. So here 50%/100%; that gives a DIC of 5.

So there are actu­ally two vari­ables that deter­mine the DIC: energy consump­tion in idle state and the percent­age of idle. Now let’s look at them sepa­rately to show the effects on the DIC through an example.

Savings Buttons

In the server hard­ware, Intel and asso­ciates have ‘baked in’ various possi­bil­i­ties to save energy when the server is in idle state. It is our expe­ri­ence that these possi­bil­i­ties are currently not used by prac­ti­cally anyone. Let’s assume that these ‘power buttons’ can save 25 to 50% when the server is idle.

In figure 2 we see the effect on the DIC. The DIC has been improved from 5 to 6 using the previ­ous example as a base. It will also be clear that the idle power consump­tion will never reach zero. After all, the CPU is always active for the back­ground tasks. It is, however, visible at the bottom that the energy saving is 18%. And that’s just because we use less energy in the idle state! 18% of the total IT load.

Reduce idle to 25%

Now let’s lower the idle percent­age from 50% to 25%. This means that the active state of the server has gone from 50% to 75%. A DIC from 5 to 8! So a huge improve­ment. However, in figure 3 it is stated: no energy savings. That’s true, but then again it’s not. The server(s) use the same amount of energy in active state and idle state. Nothing has changed in this example, so no savings are made. However, it is stated below that to process the IT activ­i­ties, only 7 servers are now needed, compared to 10 before. So savings have been made. This is because 3 servers are needed less, which is a hard­ware saving. In addi­tion, energy has also been saved, because only 7 servers are ‘on’ instead of 10. So in addi­tion to the hard­ware savings, 30% has also been saved on the elec­tric­ity bill.

What is the DIC now?

We expect that for most data centers the DIC will be 2 or less. This is due to the combi­na­tion of no acti­va­tion of the power saving options in idle state while the servers are a maximum of 20% active, so 80% idle or inac­tive — see figure 4.

Mind you, a DIC of 2 actu­ally says that 80% of the IT energy is used for NOTHING.

In the compar­i­son you see that 25 servers are now needed.

Booster

The above is not an easy fare. It all doesn’t seem logical, it is also ‘counter intu­itive’. However, it is the first attempt to ‘link’ power consump­tion of the IT systems to the real, measured power consump­tion in each data center. What it mainly shows is the enor­mous ‘waste’ of energy. The DIC has a role in raising aware­ness of energy consump­tion or waste in the data center. It is there­fore primar­ily a driver of improve­ment, because of a better score than before.

A first start

Are we there now? No not at all. It’s really just a start. I just checked how many servers there are in the Nether­lands. Between 300,000 and 500,000. Are we all going to deter­mine the idle percent­age of that? That’s quite a job. Noting and keeping track of current measure­ment per rack per month is also a chal­lenge. That is prob­a­bly more than 50,000 measurements.

And then also the UPS output current measure­ment. Those are just a few compared to racks and servers. However, if I have deter­mined the highest maximum value, is that also the maximum? Is it possi­ble that certain servers are in idle state? If they were active, wouldn’t the UPS load be higher? And that also applies to the minimum value. Who tells me that all servers are idle at that moment and not active?

Questions

Power measure­ment per rack also means that I have to average all the idle percent­ages of the servers in that rack. That leads at the very least to ques­tions and in any case to a lower accu­racy. Is it already known how the servers can be made much more active? Appar­ently not, other­wise we would have done it. Virtu­al­iza­tion is actu­ally a first step to go from less than 10% occu­pancy to as much as 20% active. What are the next steps? Do you already know them?

Role of government

And if you have read the previ­ous article: why are the data centers and IT customers going to measure and improve the DIC again? I expect they won’t. Not because they don’t want to, but because they don’t have to and have other prior­i­ties. Energy costs are not a prior­ity, so there is no real atten­tion to it now and in the future. That only happens if the govern­ment inter­venes. And who directs the govern­ment? That’s right, you and me. We will have to draw atten­tion to it. For the enor­mous savings poten­tial or as you wish: by paying atten­tion to the enor­mous waste of energy in our IT. The glass is less than half full. So far, no one has really gotten ‘fat’ about this topic. That is why it will take some time before the data centers them­selves start making DICs. Unless we manage to get the govern­ment moving. 

Mees Lodder †

Mees Lodder †

Mees Lodder werd op 14 mei 1961 geboren in Groningen. Afgestudeerd aan de TU Delft (Werktuigbouwkunde) was hij werkzaam voor onder andere Van Leer Consult, Brouwer Delft, CDS Engineering, Ballast Nedam en Bulters & Bulters.

Related

Pin It on Pinterest

Share This