New Technology Slashes Data Center Energy Consumption


20 Percent Power Savings

20 Percent Power Savings

I did a bunch of reading over the Fourth of July holiday and was sent a link for a newly funded project in Europe which claims to be the beginnings of a revolutionary new technology that will allow data center operators to save 20% off their power bill! Terrific! Now I have heard that before, but ever optimistic, I read on! Anything that promises 20% savings in Data Center power using revolutionary new technology STILL catches my attention! What’s more, if true, those power savings come with a huge savings in Carbon-Footprint at the same time. (Sure Europe is the main marketplace concerned with carbon, but WOW, we can save ‘tonnes’ of carbon too!) I had to read on…

The FIT4Green.EU project wants to address the big white elephant in the room. They are working on the energy savings area nobody wants to talk about. They even received 3 MILLION EURO funding to progress this new technology. The core concept idea for this research is based upon the following:  If you closely study the applications being serviced by the data center, and then apply some really smart logic to the control plane, you can figure out which devices can be SWITCHED OFF without affecting the overall available capacity RELATIVE TO THE REQUIRED DEMAND and thereby maintaining the required service levels. Now here’s the key: The ‘new’ technology promises to use the existing software management frameworks to DO the POWER state change, automatically, dynamically! (Read: Automation) Perhaps there really is something here….

The industry talks about dynamic capacity management, but very few companies are actually doing this at the hardware level (the place where power is consumed!). In fact, entire startups exist for this sole purpose (take a look at a few favorites of mine Vigilent and Power Assure, both in Silicon Valley). The secret sauce envisioned in this new Fit4Green project is using the existing management software framework infrastructure to manage the state control of devices that are unneeded at every point in time. Fit4Green aims to model and determine WHICH servers and other devices can be turned OFF (and subsequently back ON) at any/every point in time. Simply, model the transaction demand over time, look at which devices are in the critical data path in some type of cascaded or parallel fashion, and then dynamically tune the transaction model to account for the data center powering ON and OFF the gear that is ‘not needed’. As an example, if Black Friday in November requires 50,000 servers, then why should the operators leave all of them running at 2AM in the quiet month of August when perhaps 8-10,000 servers would do? (a huge savings). Remember that servers take TWO-THIRDS of their rated power consumption maximum before they do any actual work. To save the large amounts of power, you MUST turn them completely off (or place into deep sleep C4+/S3/etc state if you are adventurous).

Fit4Green hopes to take the next step by helping with the ANALYTICS and CONTROL portion of doing this. Frankly, the hardware required to turn a server’s power on or off is likely already in place or easily retrofitted. The existing hardware easily allows programmatic control of many of these devices to be powered on or off. IPMI and ILO type servers have had this ability for years.  And for other types of devices, intelligent switched outlet rack-based PDUs are also becoming commonplace and cost-effective. The piece of the mystery that has always been elusive and avoided in most modern data centers is the software layer of INTELLIGENCE to make the power-optimizing decision and DO IT (turn the servers off or on)!!!!

I applaud this Fit4Green effort, and am thrilled to see any project that may be able to speed the take-up of TANGIBLE dynamic power management across the industry. It is true that many of the old-school Data Center operators are still hesitant to have programmatic and automatic control of their devices’ power states, but that my friends is just a matter of time and the right motivation. Remember Apple’s original tablet (the Newton, circa 1993) failed miserably, but look at the iPad now! Times change as we become more enlightened…

Have a look here for more info on the Fit4Green project:

About admin

Mark Harris Fremont, CA
This entry was posted in Uncategorized. Bookmark the permalink.

One Response to New Technology Slashes Data Center Energy Consumption

  1. Great article – and spot on about catching the inevitable nature of ‘dynamic capacity’. I just the love the Apple Newton versus Apple iPad analogy. Enabling technology can be everything. So, timing is everything.

    As you suggest, the $64K question is the ‘INTELLIGENCE’? Hmm. What must the missing ‘INTELLIGENCE’ look like?

    You mention that we have the basic ‘control plane’ knobs to manage power via IPMI, ILO, etc. In fact, these deliver both the low-level power ‘knobs and dials’. And you mention the steady proliferation of power states (over the years), for those wishing to avoid the operational risk associated of a full power-cycle/reboot (especially on an unpredictable, automated, ‘programmatic’ fashion).

    Any wager on whether the whole IT ‘cloud’ frenzy plays an enabling role?

    For example, infrastructure-as-a-cloud (IaaS) offerings like Amazon Web Services, et. al, encourage the proliferation of ‘stateless’ servers behind a load balancer. What is it about these sorts of set-ups that sort of naturally suggest ‘dynamic capacity’? And how might ‘private’ or ‘hybrid’ clouds bring these benefits to the mainstream data centers of corporate america?

    From my point of view, mainstream data centers are full of ‘servers’ that just don’t lend themselves to this most-obvious sort of ‘dynamic capacity’ style management … up to and including ‘power down/off’. For example, when can a MS Windows file/print sharing server, or a network-attached-storage server, or even application database servers be ‘powered down’? It isn’t like corporate business unit’s are eager for IT services to pull back from todays 7x24x365 availability goals. Rather often, business continuity requirements seem to trump power savings. For example, when might a corporate email server go down just to save power?

    To the degree this is true, say for 80% of data center ‘servers’, then the putative savings evaporate pretty fast. If 20% of the load devices (i.e. ‘servers’) are subject to ‘dynamic capacity’, then only 4% of the total electricity load can be saved (… at best).

    As you suggest, the ‘INTELLIGENCE’ bit needs exploration. So, going back to timeliness, what new, enabling IT changes/developments might help? How about today’s mass adoption of the various ‘instant-on’ technologies (that reduce the time to either power-up or upgrade the power-state). This is proliferating at all levels, from CPUs/chipsets that can ‘dynamically’ power/shutdown CPU ‘cores’ to fast boot times for operating systems. What if load devices (e.g. ‘servers’) could power sip, but stay up. and then ‘wake’ (on LAN … if you will), within 1 second of receiving requests for a bunch of work.

    I’m looking forward to you next blog installment! Cheers.

Leave a Reply