Cloud vs. On-Premises – Hard Dollar Costs

Edit 2/3/2017 - For a comparison of this article versus Microsoft's TCO Calculator, see my revisit of this topic here.

This article is designed to be a little different. I assume that you already know what the cloud is and have already read the plethora of other “Cloud vs. On-Prem” whitepapers that cover costs in an ephemeral and unspecific way. I know how you feel after all of those hours of research, you understand the pros and cons at a high level, but you can’t take generalities to the bean counters, they want hard numbers. “Why can’t somebody just give me the numbers!” you have asked even though through your research, you understand the dilemma. As Denise Richards might say "It’s complicated". Of course, this actually is complicated. Pricing out technology solutions can become incredibly complex and further trying to compare two such disparate models as cloud versus on-premises makes the task seem insurmountable. Read on, I’ve solved that for you...yay me! Here are the numbers.

The Ground Rules

In order to get you the numbers, I have set out a number of ground rules in order to guide the analysis. These ground rules are:

• Simplify – When you have a complex problem to solve, you need to simplify things. Making sound assumptions and weeding through the noise and complexity is paramount and I will specifically call out where I have simplified and why.
• Apples-to-Apples – Because things can be done so much differently in the cloud versus on-premises, especially when it gets to PaaS and SaaS solutions, this rule ensures that we are really comparing the same type of solution between cloud and on-premises.
• Reality – Reality checks are important, we must be sure that we are not comparing unrealistic scenarios
• Handicap – This one may not seem so obvious, but we want to handicap the cloud so to speak. Why? Simply put, the cloud is new to a lot of folks, and even when you show up with hard numbers, they are going to argue and nitpick every little thing about the analysis. So, let’s anticipate those objections and perform our analysis in such a way that we make assumptions that favor on-premises and are unfavorable to the cloud.

Simplifying Assumptions

These simplifying assumptions have been designed to simplify the cost analysis and also to handicap the cloud, eliminating many of its purported advantages.

• You have a client computer, network switch, firewall and Internet connection and are going to have those regardless of whether the system is on-premises or in the cloud
• We will compare with a single cloud environment, Microsoft’s Azure
• We will ignore costs of setup and configuration. You will either do it yourself or pay an IT professional to configure it. One way or another, we will assume that these costs are roughly equivalent
• We will ignore tax implications (cap-ex versus op-ex).
• Even though time is money, we will ignore time to implementation
• We will assume no maintenance fees or maintenance issues, everything runs perfectly
• We assume a 3 year (36 months) lifecycle for all hardware
• We assume that all software licensing costs are included or equivalent
• A month has 744 hours
• Round numbers, for ease of the math involved and to further handicap the cloud, we will round on-premises costs down and cloud costs up.
• All calculations will be in monthly cost comparisons

First Scenario – Single Server On-Premises vs. IaaS

This is about as simple as I can make it. What are the costs of running a single server on-premises versus in the cloud in an Infrastructure as a Service (IaaS) model?

On-Premises

After eliminating the elements covered in our simplifying assumptions, the costs involved with an on-premises, single server solution can be broken down into:

• Cost of the server hardware
• Power consumption
• Space

Costs of servers can vary wildly, from about $1,000 to $30,000, depending upon its configuration. I chose a Dell R320 with a nominal cost of $2,000. $2000 / 36 months = $55.56 so our monthly cost of this server will be set at $55. We will ignore that we would likely need a separate network switch and battery backup for this and assume that these factors are included in our real estate (space) costs.

In terms of power consumption, we aren’t completely blind to reliability, our Dell R320 is configured with dual 350W power supplies. Calculating the kWh consumed by these power supplies, we use RapidTables’ Watts to kWh Calculator with 700W and 744 hours to arrive at 520.8 kWh consumed per month. We round this down to 520 kWh but, we must account for the fact that this is maximum energy consumption. Essentially, the power supplies are running flat out, which means that the processors are pegged 24x7. Reality? Not likely. So, we will assume that, on average, our power supplies are pulling 30% of their maximum power rating and that includes the fact that power supplies are only about 80% efficient. In other words, if a power supply needs to deliver 100W and is 80% efficient, it will actually burn 120-125W. 520 * .30 = 156 kWh. However, we must also insert reality in terms of Power Usage Effectiveness (PUE). PUE is the measure of how effectively power is delivered to servers versus being used to power heating and cooling systems, etc. A recent survey says that PUE averages about 1.7. So, 156 * 1.7 = 265.2, which we will round down to 265. Finally, we must convert this to dollars. Government websites indicate that the commercial price per kWh in the United States averages about 11 cents. So, 265 * $.11 = $29.15, which we will round down to $29.

The server has to be located somewhere. You may stick it in a closet or in a data center, but at some point, there is cost associated with where it is located. To simplify the myriad possibilities of where the on-premises server is located, we will simply assume a cost of $10 per square foot per month. This is not entirely a number picked from thin air. We will assume that this cost per square foot covers having the server take up space, network and potential power outages. Let’s say it is running in some data center that provides all of that for $10 per square foot per month. The Dell R320 is approximately 2 feet wide by 4 feet long, so 8 square feet, gives us $10 * 8 = $80.

So, at the end of the day, you have a server sitting in a data center on the floor which you are effectively paying $55 (capital) + $29 (power) + $80 (space) = $164 per month.

Cloud

Cloud cost calculations are simple in comparison to on-premises. A roughly equivalent server in Microsoft Azure IaaS, a D2 SSD 2 cores 7 GB RAM 100 GB disk in Azure will cost you somewhere between $193.44 per month to $208.32 per month depending on the data center. We will simplify this to $200 per month. This includes the server, power consumption and space.

Analysis

Through simple math, $200 (cloud) / $164 (on-premises), we can easily calculate that cloud is approximately 20% more expensive than on-premises for this single server scenario. It is left to the reader as to whether the advantages of the cloud are worth that extra 20%. The advantages being:

• Time to implementation – Provisioning a Azure IaaS server takes about 15 minutes to have a fully functioning server, complete with OS installed and configured versus at least days for on-premises since the server needs to be ordered, shipped, installed and configured.
• Net Present Value – That $2,000 capital expenditure could be put to better use investing and earning interest, for example. Thus, that $2,000 capital expense technically costs you more than $2,000 because that money could have been invested and earning interest over the 3 year lifecycle of the system.
• Tax Implications – The hardware bought as a capital expense must be depreciated over time while cloud solutions are considered operational expenses that are deducted as costs.
• Flexibility – If, in a year’s time, it is discovered that you no longer need the server, in the cloud you simply turn it off and no longer pay for it. With on-premises you are essentially stuck with the server and can try to recoup some of the capital expense by selling it, which; of course, you would pay taxes on that income.
• Ongoing Costs – If the lifecycle for the hardware is 3 years, then technically the costs to migrate the system to something new should be accounted for in the total cost of the on-premises solution. In the cloud, if Microsoft needs to upgrade its servers, that’s not your problem.
• Reliability – Microsoft guarantees 99.9% uptime while with on-premises you probably don’t have a guarantee at all.
• Security – Microsoft assumes breach within its security infrastructure, meaning that they are constantly, actively looking for breaches in security. Microsoft is constantly running “red team/blue team” white hat hackers, one team actively looking for penetration points and the other team actively looking to stop them.

Second Scenario – Multiple Servers On-Premises vs. IaaS

Sure, I know what you are thinking, that first scenario was perhaps a little too simple. What is really bugging you is that $80 space cost of the on-premises server, nearly half the cost! Where is the reality I promised? It’s just not realistic that you would only have one server in that 8 square feet, you would have a whole rack of servers in that space.

Fair enough, let’s do the math.

On-Premises

We’ve already done the heavy calculations for on-premises, all we need to add now is a 42U rack unit, which will run you about $1,000. $1,000 / 36 months = $27.78 per month which we will round down to $27/month. So, for the on-premises solution, we have (assuming we maximize the use of the rack with 42 1U servers):

$27 (rack) + 42*$55 (servers) + 42*$29 (power) + $80 (space)

or

$27 (rack) + $2,310 (servers) + $1,218 (power) + $80 (space) = $3,635 per month

Yes, in today’s world you would likely have fewer servers and run a few big host servers that run 42 virtual machines, but those few big servers are going to have bigger power supplies, run at a higher power consumption percentage and you need to pay for the virtualization software, added complexity, etc. So, keeping it simple, let’s just say that this scenario and that alternative scenario are roughly the same cost per month.

Cloud

42 * $200 = $8,400 per month

Analysis

Aha! $8,400 / $3,635 = 2.31 or that it is roughly 230% more costly to run cloud than on-premises. There you go, you knew it all along, the cloud is WAY more expensive than on-premises! Perhaps. But here is where we need to inject a little bit of reality in the other direction.

First, with 42 servers to manage, you are likely going to need something to view and manage them with. In Azure, management software giving you a single pane of glass to monitor and manage your servers is included. In addition, this management software allows you to schedule and automate tasks such as stopping and starting servers, backup VM’s, provision systems, send emails, etc. For all of that, remember apples to apples, you are easily looking at spending at least $2,000 in software. So, add another $55 of capital software expense.

With 42 servers, you are going to need load balancing. It is highly likely that you have some web servers in there and other types of servers that need load balancing. For redundancy, you will need at least 2 load balancers. Add another capital expense of at least $10,000 or $275 per month. You’ve already filled up your rack with servers, so add another $27 for a second rack and another $80 per month for the space for that rack. We will give you a break though and not even consider the additional power consumption of these load balancers.

Peanuts right?

2*$27 (racks) + $55 (software capital) + $275 (load balancing capital) +$2,310 (server capital) + $1,218 (power) + 2* $80 (space)

or

$54 (racks) + $55 (software) + $275 (load balancing) + $2,310 (servers) + $1,218 (power) + $160 (space) = $4,072

So we’ve knocked it down that the cloud is only twice as expensive as on-premises. Big deal.

Well, one last little bit of reality to interject. With 42 servers in the mix, the odds of being able to run them all with local storage is unlikely. In today’s modern IT world, you need shared storage. On-premises, this is handled by a SAN while in Azure, this is handled by a storage account.

Let’s first look at the cloud, we can provision a 1,000 GB (1 TB) storage account for $0.10 per GB. No, that decimal is not in the wrong place. What’s more, this $0.10 buys us GRS storage (Globally Redundant Storage). What is GRS? GRS essentially means that each GB of data is actually stored 3 times locally and 3 times globally. So, essentially, each transaction is written to 3 storage nodes in the local data center and then are also written to another data center where it is also written to 3 storage nodes in that data center. $0.10 per GB to effectively have 6 copies of your data at all times with geographic separation included. 1,000 * $0.10 = $100 per month. So, for the cloud we have:

$8,400 (servers) + $100 (storage) = $8,500 per month

For on-premises, to get an apples-to-apples comparison, we are going to need 2 SANS in separate geographic locations with dedicated high-speed network connections between them. We will need 3 TBs of storage for each SAN giving us a total capacity of 6 TB. Overkill? OK, we will knock this down to 4 TB total, 2 TB for each SAN so that each has a “live” copy and a backup copy. But how do you calculate out the costs for a SAN? While we could assume away IT resources being equivalent between cloud and on-premises for something simple as servers, we can’t make that same assumption for SANs and still claim realism. Typical things that go into a SAN pricing would be the enterprise SAN arrays and software, backup systems, high speed network to connect them, 2-4 people’s salaries, power consumption, space, etc. Let’s simplify this. If you do your research, you’ll find that SAN costs per GB per month typically fall in the range of $5-$15 per GB with some people claiming they have been charged by internal IT up to $30 per GB. Let’s stack the deck in favor of on-premises. Let’s say that, for today only, you can have a geo-redundant SAN with power, space and management costs included for the low, low price of a mere $2 per GB, twice as low as any estimation of storage price that you will likely find out there today. 4,000 GB * $2 = $8,000 per month.

So, now we have:

$8,000 (storage) + $54 (racks) + $55 (software) + $275 (load balancing) + $2,310 (servers) + $1,218 (power) + $160 (space) = $12,072

Let’s give on-premises yet another break, we’ll call it $12,000 even.

$8,500 / $12,000 = .708 or that now, in a fully realistic scenario, cloud is approximately 30% cheaper than on-premises.

Conclusion

Calculating hard dollar cost comparisons between cloud IaaS versus on-premises infrastructure is complex, even with simplifying assumptions and is dependent upon the specific solution being deployed. While it varies wildly depending upon the specifics of the system being deployed, for midmarket/enterprise businesses, cloud based solutions; despite all conceivable advantages being given to on-premises, represent a potential cost savings of 30%. These are hard dollar cost savings that do not even factor in the litany of other cloud advantages cited in this paper. The majority of this cost savings is accounted for by storage costs, which are at least an order of magnitude less expensive in the cloud versus on-premises.