During last week’s DCD>LondonDay One keynote, Dean Nelson, head of Uber Compute and founder of Infrastructure Masons, shared some fascinating insights on how best to address explosive business growth. In detailing his infrastructure vision for Uber,Nelson spoke of the need to embrace what he calls a “tripod strategy” in the data center to scale compute capacity dynamically.
His recommended approach for accommodating hyper-scale growth is the triad of on-premises data centers, co-location facilities and public cloud. According to Nelson, Uber completed 10 billion trips in eight years, five billion of which took place in the last year. The expectation is that the ride-hailing juggernaut is just getting started on its meteoric rise.
“The thing that has kept me up for the last two years is keeping up with this demand,” Nelson told the attendees. All too often, we hear a similar refrain from IT leaders from all regions and industries who are struggling to stay ahead of the growth curve while facing unrelenting budget challenges.
For on-prem deployments, Uber relies on a standardized model called Uber Metal, which encompasses 576 racks of compute, database, storage and GPUs for machine learning. One concept Nelson is considering involves defragging the data center, so Uber can take advantage of the fabric architecture and move things around to avoid stranded power.
This is an interesting strategy, especially when it comes to leveraging unused power capacity for increasing energy utilization as well as reallocating power on-demand to support dynamic IT workloads. Virtual Power Systems,creators of a new category called Software Defined Power (SDP), is poised to do for power what VMware did for server virtualization.
Using a similar approach, SDP provides a common control plane for power components. As a result, SDP offers a triple-play power solution that provisions power in real time across enterprise data center, co-lo and cloud environments. Using machine learning and predictive analytics, VPS’ Intelligent Control of Energy® (ICE) platform automatically reallocates power according to capacity and availability demands—a big deal for companies of all sizes.
According to a report by Ponemon Institute on behalf of Emerson Network Power, power and cooling costs together account for 40% or more of data centers’ annual costs. Environments with power-intensive hypercompute and machine learning applications further escalate power costs. By embracing SDP, companies don’t have to overprovision power. Instead, they can readily tap into unused or backup power capacity as needed.
This approach is similar to Nelson’s defrag idea, where SDP can be utilized to create a readily available virtual power pool that can be used on demand, and when needed anywhere in the data center.
With the holiday season approaching, I suspect that Uber is bracing for another quantum capacity leap. This is where VPS’ triple-play solution will work seamlessly across Uber’s “tripod” environment to reduce overall power-capacity pressures. Having a solution like SDP, which is uniquely poised to unlock stranded power and improve energy utilization, can help IT leaders everywhere sleep better at night.