Monday, June 13, 2016

Not Another xUE Metric!

By David Gross

This industry loves efficiency metrics.  Power, space, cooling, disk, I/O, memory - whether for capacity or environmental reasons, you can spend all day running ratios to measure well a facility and the hardware inside its walls utilize their resources.  But how do you know if you're measuring the right things?


I've never found all the "xUE" metrics that get heavily covered in the press to be that useful from a business standpoint.   Moreover, while calculating basic utilization metrics isn't that difficult, they often need to be incorporated into larger equations to be useful as management targets.   To capture this, one metric I've come up with is Weighted Average Cost per Utilized kW.    To calculate this, I take Full Service Monthly Rent/Consumed kW.   It can be measured across a pod, a facility, or for an entire portfolio.


I use full service rent because it not only captures all monthly costs, but it makes modified gross leases, which include management charges, comparable to triple net leases, which charge those fees separately.   To be even more complete, you can add depreciation on internally built cabling and rack infrastructure to the numerator in order to make wholesale leases more comparable to retail leases.


You'll notice that this assumes power is the scarce resource, because the equation above makes no reference to space.   While there has always been, and will likely always be, debate about power vs. space as the scarce resource in data center builds, I tend to focus on power. The reason for this is 2/3rds of the capex for a typical REIT-built data center goes to power and cooling infrastructure.   Meanwhile, land is typically just 5%.  Therefore, the scarce resource is power and cooling equipment, not space itself, which is why wholesale providers usually price leases by the kW, and not the square foot.  Some exceptions to this, especially in downtown carrier hotels, but leasing a rack at 111 8th Ave is still far cheaper than leasing an entire room in Ashburn.


So what's a good Weighted Average Cost per Utilized kW?  (or to be cheesy, WACU)


Depends on country mix, your company size, and if you've just taken on a large addition, but generally anything over $300 suggests you're overpaying, you're not big enough for wholesale leases, you're underutilized, or you've got a large footprint outside the U.S.   This also assumes Tier 3/4 facilities.  If you're going for lower tiers, or less redundancy, the meaning starts to get lost.  


Lastly, why measure this?  


Primarily because power, space, and cabling/rack capex can consume 20-40% of an IT or cloud budget, they require long-term leases, and aren't subject to Moore's Law unit cost reductions like compute cores, RAM, I/O, network, or disk.  By measuring cost and utilization in one equation, you can get a quick view of how efficiently you're using your data center dollars.

Tuesday, June 7, 2016

Data Center Stocks Close at Record Highs

By David Gross

Continued strength in bonds and investor giddiness about data centers helped push DLR, DFT, COR, CONE, EQIX, and QTS to record highs today.   Except for EQIX, all were up over 2% today, ahead of the 0.53% gain posted for the NAREIT Composite Index, which covers REITs for all asset categories. Outside the REIT category, INXN also climbed nearly 2% to close at 38.79.


Outside of yesterday's post-close announcement that QTS was buying DFT's NJ1 property for a fairly low $7,000 per available kW, not much news behind today's surge. Still, the combination of possible rate hike delays and Wall Street analysts catching on to a trend long after it's started aren't hurting.  Wells Fargo, for example, initiated DLR last week with an "outperform" rating.   Of course, you can't call your broker and ask to outperform the stock, and even if you could, where was Wells in 2013 when Digital was at 45 yielding 8%?