SaaS is going to come out ahead of any reasonably calculation of carbon emissions versus on-prem. Multi-tenancy is just a lot more efficient. Look at the data centers of companies like Google, Amazon, and Facebook. Most corporates wish they could come close as they watch these companies dictate every detail right down to the exact design of servers in order to minimize their costs. As everyone agrees, most of that cost is energy.
So choose SaaS if you’re worried about carbon, and yes, it could become another axis of competition in terms of exactly which Cloud provider does it best.
Tom Raftery immediately responded:
The answer is that it depends, tbh. It depends entirely on the carbon intensity of the data centre (where it sources its energy), not the efficiency of the data centre.
If you have a data centre with a PUE of 1.2, and it is 50% coal powered (not atypical in North America, Germany, Poland, and others, for example), it will have a higher CO2 footprint than a data centre with a PUE of 3.0 powered primarily by renewables – again I have run the numbers on this and published them.
Similarly with on-prem. If I have an app that I’m running in-house, and I’m based in a country like Spain, France, Denmark, or any other country with where the electricity has a low carbon intensity; then moving to the cloud would likely increase the CO2 footprint of my application. Especially if the cloud provider is based in the US which has 45% of its electricity generated from coal.
Tom is the chief analyst for Greenmonk, which writes about this sort of thing for a living. He’s been quoted by others who are in the same camp such as Brian Proffitt on ReadWriteWeb. And who wouldn’t love a nice juicy story to put those darned Cloud vendors in their place? Those guys have been riding high for too long and ought to be brought down a notch or two, harrumph.
I have a lot of problems with this kind of math–it just doesn’t tell the whole story.
First, I can’t imagine why Tom wants to be on record as saying that PUE (Power Usage Efficiency) just doesn’t matter. Sure, he has found some examples where CO2 footprint overwhelmed PUE, but to say the answer depends entirely (his word) on the sources of the data center’s energy and not on the efficiency of the data center just seems silly to me. Are there no data centers anywhere in the world at all where PUE matters? Did all the Cloud data centers with great PUE just magically get situated where the carbon footprints are lousy enough that PUE can’t matter?
I’m very skeptical that could be the case. You must consider both PUE and CO2 per Kilowatt Hour, how could we not when we’re talking per Kilowatt hour and PUE determines how many Kilowatts are required?
Here’s another one to think about. If this whole PUE/CO2 thing matters enough to affect the economics of a Cloud Vendor, we should expect them to build data centers in regions with better CO2 energy. Since they put up new data centers constantly, that’s not going to take them very long at all. Some are talking about adding solar to existing facilities as well. Now, do we want to lay odds that corporate data centers are going to be rebuilt and applications transferred as quickly for the same reasons? If you’re running corporate IT and you have a choice of selecting a Cloud Data Center with better numbers or building out a new data center yourself, which one will get you the results faster? And remember, once we are comparing Apples to Apples on CO2, those Cloud vendors’ unnaturally low PUE’s are going to start to haunt you even more as they run with fewer Kilowatt Hours.
Multitenancy Trumps all this PUE and CO2 Talk
But there’s a bigger problem here in that all data centers are not equal in another much more important way than either PUE or fuel source CO2 footprints. That problem is multitenancy. In fact, what we really want to know is CO2 emissions per seat served–that’s the solution everyone is buying. Data centers get built in order to deliver seats of some application or another, they’re a means to an end, and delivering seats is that end. The capacity they need to have, the number and type of servers, and hence the ultimate kilowatts consumed and carbon footprint produced is a function of seats. Anyone looking purely at data centers and not seats served is not seeing the whole picture. After all, if I run a corporation that has a datacenter, it’s fair to charge the carbon from that datacenter against my corporation. But if I am subscribing to some number of seats of some Cloud application, I should only be charged the carbon footprint needed to deliver just those seats. Why would I pay the carbon footprint needed to deliver seats to unrelated organizations? I wouldn’t.
Corporate data centers have been doing better over time with virtualization at being more efficient. They get a lot more seats onto a server than they used to. The days of having a separate collection of hardware for each app are gone except for the very most intensive apps. But that efficiency pales in comparison to true multitenancy. If you wonder why, read my signature article about it. I’ll run it down quickly here too.
Consider using virtual machines to run 10 apps. Through the magic of the VM, we can install 10 copies of the OS, 10 copies of the Database Server, and 10 copies of the app. Voila, we can run it all on one machine instead of 10. That’s pretty cool! Now what does Multitenancy do that the VM’s have to compete with? Let’s try an example where we’re trying to host the same software for 10 companies using VM’s. We do as mentioned and install the 10 copies of each kind of software and now we can host 10 tenants. But, with multitenancy, we install 1 copy of the OS, 1 copy of the Database, and 1 copy of the app. Then we run all 10 users in the same app. In fact, with the savings we get from not having to run all the VM’s, we can actually hose more like 1000 tenants versus 10.
But it gets better. With the Virtual Machine solution, we will need to make sure each VM has enough resources to support the peak usage loads that will be encountered. There’s not really a great way to “flex” our usage. With Multitenancy, we need to have a machine that supports the peak loads of the tenants at any moment in time on the system. We can chose to bring capacity on and off line at will, and in fact, that’s our business. For a given and very large number of seats, larger than most any single corporate application for most corporations, would we rather bet the corporation can be more efficient with on-prem software in its wholly owned data center or that the SaaS vendor will pull off far greater efficiency given that its software is purpose-built to do so? My bet is on the SaaS vendor, and not by a little, but by a lot. The SaaS vendor will beat the corporate by a minimum of 10-20x and more likely 100x on this metric. You only have to look to the financials of a SaaS company to see this. Their cost to deliver service is a very small part of their overall expenses yet most SaaS apps represent a considerable savings over the cost of On-Prem even though they carry the cost of delivering the service which the On-Prem vendor does not.
Raftery says, “energy use” and “emissions produced” have been conflated to mean the same thing. I say he’s absolutely right about that but hasn’t seen the bigger picture that it is not energy use nor emissions produced in isolation, it’s seats delivered per emissions produced. Itt’s having the flexibility to make a difference rapidly. And that is why we bet on the Cloud when it comes to being Green.
(Cross-posted @ SmoothSpan Blog)