Ready or not (and most aren’t), power density in the rack is going up, and not incrementally over ten years, but dramatically over three to five years. Can your internal data center(s) support that? Can your partners support it? My rough estimate tells me that if an average of 10kW per rack was required, fewer than 10% of data centers in operation today could handle it.
Why the Screaming?
There are a confluence of events occurring that are driving infrastructure design towards more density, and I don’t see anything reversing that trend anytime soon.
Reason 1: Converged infrastructure (more gear in a smaller space)
- Right now, converged infrastructure sales are going through the roof. Whether you’re buying from VCE, HP, Dell, NetApp, Nutanix or SimpliVity, the story is the same…more power Scotty!
- UCS Chassis – 2.0 kW – 6 chassis per cabinet = 12kW
- HP Matrix Chassis – 6.0 kW – 4 chassis per cabinet = 24kW
- Neither of the above examples are configured at max potential power consumption.
Related blog “Future Proofing”; http://www.switchscribe.com/data-center-trends-part-ii-of-iv-a-future-proofed-decision/
Reason 2: Open Compute and or HPC Oriented Infrastructure designs
- Modern infrastructure is taking on more of the characteristics of High Performance Compute designs. There are a number of reasons for this change towards HPC, but one of the main drivers is the increasing use of big data and reliance on big data as part of critical infrastructure/application environments. Simply put, there are real performance gains associated with putting more compute in smaller spaces right up next to your disk farms.
Reason 3: Cost & Space – Sustainability
- I’ve always liked density because I’m an efficiency guy. There are some obvious benefits to having your gear packed in to smaller spaces: better overall performance (efficiency) and less floor space, cabling, and racks (efficiency and cost). There’s also the long term sustainability (blog) There are estimates being made today that suggest cloud and IoT will drive our global IT power consumption footprint from 3% to as much as 20% of total power generated. If we continue to populate the world with low density data centers you’ll soon be able to walk on top of them all the way across the country.
Reason 4: Containers
- A key push behind the use of containers after ease of deployment, etc. is efficient use of infrastructure. A big part of this efficient use of infrastructure comes from a much higher average utilization of the CPU. This higher utilization will inevitably result in greater power use per rack unit.
Reason 5: Moore’s Law slow down
- It might seem weird to think that a slowdown in Moore’s law would equate to higher power density, but I believe it does for at least two reasons. Disk and Networking continue to advance at Moore’s law or faster. This advance will mean: 1) More CPUs/Servers will be needed per square foot to support bigger and faster disks and fatter, speedier networks, 2) we’re less likely to see that same decreases in power draw on the CPU as a result of the reduced iterations of x86 chips.
Related blog “Not Your Daddy’s Data Center”: http://www.switchscribe.com/not-your-daddies-data-center-nyddc/
What am I getting at?
I expect that over the next three to five years many data center owners will discover that their facilities won’t support what their IT staffs are demanding; and even worse, many of the providers you speak to won’t be able to support your need either. Here at the SUPERNAP, we’re seeing the demand for density increase with a large percentage of our customers. We’re already doing thousands of racks at over 15kW delivered and many that are much denser with the trend accelerating.
Professionally Copy Edited by @Imkestinamarie (Kestine Thiele)