Like most others responsible of hosting large quantities of data, I have been engaged in numerous debates regarding cloud storage versus hosted storage. Is cloud storage secure? Is it reliable? Is it HIPAA compliant? If we are a global enterprise, do we know where the data is stored? This debate will rage on until standards are established and cloud storage can demonstrate that it is enterprise ready. These debates are thoroughly discussed and each side defended in trade publications and blogs throughout the web. However, it is clear that cloud services are here to stay.
What appears to be discussed less are the cost considerations. I have recently been involved in a detailed cost analysis of cloud storage vs. hosted (co-location) storage. My client had done the math, and they clearly demonstrated that cloud storage costs approximately 5x more than hosted storage. Unfortunately, analysis shows that the cost difference is not so clear.
Disclaimer
First, I make no warranties that my analysis will model your real world experience or make you rich and better looking. Second, I welcome anyone to challenge or correct my analysis. Not because I am sure it is correct, but rather because this is a simplified analysis using some numbers (e.g. server pricing) provided by other people. The spreadsheet used in this analysis is available at the end. My models are often detailed and (overly) complex and sometimes difficult to comprehend. However, they also tend to be revealing and defensible. Unfortunately, this analysis may be neither. For brevity this analysis is simplified, and some items were excluded or assumed in the name of simplicity. I will do my best to try to identify these abbreviations.
Simple Comparison
The typical back-of-the-napkin analysis is fairly straight forward. Let's assume that we need to storage 100TB of data with a certain level of redundancy. For the hosting side, we used a HP DL320 with 12 x 750GB drives. At a (client) estimated cost of $7,500, this server has an effective storage capacity after planning some redundancy of 7.8TB. At this capacity, it will take 13 of these servers to host 100TB of storage. Figuring a 3 year useful life for these servers and a 9% cost of capital, these servers have an amortized monthly cost of $3,326.74. Converting this to a common cost per GB, this hosting set-up costs $.033/GB. For the cloud side, we used Amazon whose pricing is published, and their pricing is roughly in line with other cloud storage providers. Actually, other storage providers are trying to put their pricing in line with Amazon to try to compete with the giant. Amazon charges $.15/GB for the first 50TB and $.14/GB for 50-400TB. This gives us a net monthly cost of $.145/GB.
This simple analysis shows that managed cloud storage costs 4.4 times more than hosted storage. The analysis is admittedly not complete, but the omissions are relatively small, and the difference is so great ($522,000 vs $119,793 over 3 years) that this is a slam dunk in favor of hosted storage. This is often as a complete as financial analysis goes on this matter, and unfortunately it is often wrong.
Considerations
The above analysis misses some significant and very real considerations. Depending on your situation, some of these considerations may apply and some may not. Again, the spreadsheet is included at the end, and you can tweak it to your heart's content.
Hold Your Cash
Cash is King. It always has been, but even more so today. If you are a Fortune 500 company, you may not care as much about cash flow, especially the amount of flow that we are talking about. Otherwise, hold your cash. In our current economic situation, venture funding and bank loans are virtually non-existent for small companies. In the above analysis, unless your company is solid enough to get a capital lease on the equipment, the hosted solution will drain your bank account of over $105K in month 1. Even with a loan or lease, your will have a sizeable liability hung onto your balance sheet for 3 years. Cloud storage on the other hand will cost you about $14,500 in month 1. If you are an early stage company, what is the best way to spend that other $84K? On big iron in a data center or on marketing and product development?
Opportunity Cost
Opportunity cost can be measured in several ways. Let's first consider the financial opportunity cost. As we just mentioned, purchasing hosted storage will likely cost you out of pocket cash, and that cash will not be used for other business objectives. The money will not be used to pay down debt, further product development or help land new customers. Purchasing the servers may still be the right business decision, but be aware there is a real opportunity cost.
Second, let's consider the market opportunity cost. One way to reduce your out of pocket cash is to not spend it all at once. Unless you are storing existing data, if you need 100TB over 3 years, you likely don't need 100TB today. Buy what you need today, and when you need more, buy more. A buy-as-you-need-it strategy makes logical sense, but it has problems in the real world. First, in many corporate environments, projects get funded when a project plan is reviewed and approved by upper management or a steering committee. In this case, you better get all the money you need for the project, because it can be very difficult to go back to the well a second time. And if you do use a buy-as-you-need-it strategy, how do you know when you need more storage? You are in an enviable business environment if your capacity consumption is smooth and predictable. If you are serving customers, what happens if a promotion is unexpectedly successful and customers flood your service? How long will it take to get new storage funded, ordered, delivered, installed, tested, and turned up? How many customers could you lose during that period? Who will be held responsible for this market opportunity cost?
Opportunity Risk
Closely related to opportunity cost is opportunity risk. How much will you spend to minimize the opportunity cost? The first way to minimize your risk is to assure that you have sufficient spare capacity to absorb business and market fluctuations. Do you assure that you have 20% spare capacity? 5TB of spare capacity? Even the full 100TB may turn out to be not enough storage. This is one of the direct failings in our simple analysis above, there is no spare capacity. With cloud storage, you pay for what you use when you use it. Adding spare capacity to the hosted solution will increase the effective cost of the hosted solution.
Spare capacity is also idle capacity, and I have seen idle capacity kill many large projects. The first scenario for idle capacity is classic, and it happens over and over again. Marketing wants to launch a new product, they are projecting wildly successful sales. because no one has a product like this. The tech team may doubt the sales projections, but they are focused on making sure that the software works correctly and is delivered on time. Besides, its marketing's responsibility to meet their numbers, and we don't want to go the project funding well twice. So the full 100TB (plus 20%) spare capacity is purchased and installed ready for eager customers. However, the product launch is delayed a month to coincide with a major trade show. Customers are excited, but the sales cycle turns out to be 6 months instead of 6 weeks. And then 4 competitors announce slightly more advanced products in the same category. After a year, sales are starting to ramp-up with version 2.0, and the future is starting to look bright. However, with only 10TB used, the project has continued to accumulate accounting loses of over $3.3K/month (just for the storage), and this accounting burden continues to burden the financial resources of the project. Marketing budgets are tightened, sales slump and management decides that the product was not the home run they had hoped for. With cloud computing you pay for what you need. If you are only using 10TB, you only pay for 10TB, and (hopefully) you have revenue to offset all your storage costs.
The second consideration associated with idle capacity is a little less obvious. In the simple analysis, the cost of the hosted storage was made low by amortizing the cost over the 3-year life of the equipment. This is valid if the equipment is used for 3 years and only 3 years. Using amortization, a business is committed to paying for the storage capacity for the useful life of the equipment. Many software products last longer than 3 years, but some do not. And if storage is purchased 2 years into a project, the amortization last until year 5. Think about your environment, or even more dramatically, the technology marketplace. How much has changed since 2006? Do you even look at technical documents from 3 years ago? If our sample project ends 1 year later, our amortized monthly cost is now $.092/GB rather than $.033/GB. With cloud storage, you not only pay for storage you use, you only pay for it while you are using it.
Moore's Law
Moore's Law roughly states that the complexity of integrated circuits for minimal cost will double every 2 years. More generically, Moore's Law has been applied to describe both the exponential increase in computing power and the corollary decrease in the cost of a unit of computing power. The concept of Moore's Law also applies to data storage. Through anecdotal evidence, I estimate that the cost of a GB of storage drops in half roughly every 2.5 years, and I see no evidence of this trend ending in the near future. I encourage someone to provide a detailed analysis of this theory, but that analysis is beyond the scope of this article. Moore's Law will come into our more complete analysis below. In simple terms, if you believe that Moore's Law will affect the cost of storage in the future, use the Moore's Law numbers where available. If not, then simply ignore the Moore's Law modified numbers.
Revised Analysis
I have revised the simple analysis above to include some of these considerations. As mentioned at the beginning, even this revised analysis is simplified. Overall, this analysis probably gives a little advantage to hosted storage, because some of the excluded items would likely cost more for hosted storage than for cloud storage.
The numbers discussed here do not include bandwidth, application servers, web servers, or other computing resources. All of these would likely be required for any solution whether cloud storage or hosted storage is used. You definitely do not want to use hosted computing resources and cloud storage resources or visa versa. Network bandwidth within a cloud provider and within a hosting site is free. But bandwidth in, out and between each facility can get expensive. Support and administration expenses have also been excluded with the assumption that these functions would be needed for both cloud storage and hosted storage. Although, I can imagine that these expenses would be a little larger for a hosted solution. Also, this analysis excludes cabinets, cables, power, monitors, etc. These expense items are only required for hosted storage. In the interest of keeping things simple, these costs were considered relatively small and shared among storage and computing resources. However, if you are paying $3,000/month to host a cabinet of storage at a hosting facility, your storage costs have almost doubled. This scenario also excludes inflation, since inflation has been negligible over the last several years. Given the massive global influx of capital into the world markets, inflation may become and issue in coming years.
I will look a three scenarios below: single instance ramp-up, linear ramp-up, and periodic wave ramp-up. The reality of how storage consumption ramps up is probably more elaborate than these, but I think you will find these complex enough.
Single Instance Ramp-Up
This is our simple back-of-the-napkin analysis from above (i.e. We need 100TB that will be purchased and consumed in month 1, and we will provide this storage for 3 years). It is included here to provide reference numbers, and as shown above, the numbers will heavily favor a hosted solution. If you are installing a fixed amount of storage, consuming it all almost immediately and continuing to provide the storage for the life of the equipment, use hosted storage. It will be much less expensive. As shown by the numbers, if there is no need for flexibility in storage consumption and no need for spare capacity, this is likely your financial choice. If any of these conditions are not met, Hosted Storage will likely not achieve numbers anywhere close to this chart. For those who are not able to predict the future quite as clearly, please read on.
| Single Instance Ramp-Up | Cloud | Hosted |
| Out of Pocket - Month 1 (purchase) | N/A | $105,400 |
| Out of Pocket - Month 1 (lease/loan) | $14,500 | $3,327 |
| Out of Pocket - Year 1 (lease/loan) | $174,000 | $39,921 |
| Out of Pocket - 3 yr. (lease/loan) | $522,000 | $119,763 |
| Cost/GB - Month 1 | $0.145 | $0.033 |
| Cost/GB - Year 1 | $0.145 | $0.033 |
| Cost/GB - 3 yr. | $0.145 | $0.033 |
Linear Ramp-Up
The linear ramp-up model, as the name implies, assumes no storage is consumed on day 1, and all 100TB storage is consumed at the end of 3 years. In between, the amount of consumed storage increases an equal amount (100TB / 36 months) each month. This scenario also introduces some new variations. First is the spare storage limit. When calculating expenses in this scenario, we assume a buy-it-when-you-need it strategy is used. If not, go back to the simple scenario and pay for all of the storage from day 1 regardless of when it is consumed. The numbers consider a 20% spare storage threshold. As storage consumption increases and spare capacity drops below this thresholds, new storage is purchased. The second variation considers Moore's Law. Since this scenario assumes buying storage over time, this scenario estimates a lower cost of storage purchased in the future.
| Linear Ramp-Up | Cloud | Cloud (Moore's) | Hosted + 20% Spare | Hosted (Moore's) + 20% Spare |
| Out of Pocket - Month 1 (purchase) | N/A | N/A | $15,400 | $15,400 |
| Out of Pocket - Month 1(lease/loan) | $417 | $417 | $486 | $486 |
| Out of Pocket - Year 1(lease/loan) | $32,500 | $22,538 | $12,224 | $11,244 |
| Out of Pocket - 3 yr.(lease/loan) | $272,750 | $122,394 | $80,230 | $62,225 |
| Cost/GB - Month 1 | $0.145 | $0.150 | $0.175 | $0.175 |
| Cost/GB - Year 1 | $0.145 | $0.090 | $0.050 | $0.043 |
| Cost/GB - 3 yr. | $0.145 | $0.081 | $0.040 | $0.027 |
Periodic Wave Ramp-Up
The linear ramp-up scenario attempts to add some real life variation to our analysis, however, business is rarely linear. The periodic wave, like the linear ramp-up, increases storage consumption over time. In an attempt to introduce some real world variation, the rate of increase is modified by a periodic wave (e.g. some months a big increase, some months no increase or even a decrease). This scenario also considers spare capacity thresholds and Moore's Law.
| Periodic Ramp-Up | Cloud | Cloud (Moore's) | Hosted - 20% Spare | Hosted (Moore's) - 20% Spare |
| Out of Pocket - Month 1 (purchase) | N/A | N/A | $22,900 | $22,900 |
| Out of Pocket - Month 1(lease/loan) | $1,491 | $1,491 | $723 | $723 |
| Out of Pocket - Year 1(lease/loan) | $52,523 | $38,010 | $19,799 | $18,739 |
| Out of Pocket - 3 yr.(lease/loan) | $280,476 | $122,394 | $92,303 | $77,695 |
| Cost/GB - Month 1 | $0.145 | $0.150 | $0.073 | $0.073 |
| Cost/GB - Year 1 | $0.145 | $0.090 | $0.041 | $0.038 |
| Cost/GB - 3 yr. | $0.145 | $0.081 | $0.040 | $0.029 |
The Conclusion
Simple Financial Conclusion
In almost all cases, hosted storage is cheaper than cloud storage. When viewed on a cost/GB basis, cloud storage was typical 2 - 4 times more expensive once storage consumption gets up to scale. When hosted storage is purchased through a bank loan or equipment lease, the out of pocket cash drain is generally lower for hosted storage than for cloud storage. However, when purchasing the equipment outright, hosted storage creates a much larger cash drain in month 1 than cloud storage, obviously. What might be less obvious is the on-going cost of the lease/loan option. This analysis stops after 3 years. However, the ramp-up solutions continue to pay for the equipment from the date of purchase. For some equipment, this may be an additional 35 months. These hidden future payments are the reason why the 3 year out of pocket expense for a ramped up storage plus spare capacity is lower than the simple scenario without spare capacity.
Best Business Conclusion
The Best Business Conclusion would of course include discussions of security, data integrity and application compatibility. However, this is a financial discussion, so we will limit our "best" business conclusion to the financial issues discuss here. The 3 year out of pocket expense for cloud storage for the ramped-up scenarios, even when Moore's Law is considered, is greater than the out of pocket expense of purchasing 100TB of storage straight out. But what happens if only 10TB are used over 3 years? The cloud storage solution would cost much less. And what if the storage was only needed for 3 years? Then cloud storage payments would stop when the need for storage stopped. The lease/loan payments would continue for the useful life of each piece of purchased equipment. Even if the equipment is purchased out right, depreciation expense would continue for some time.
The decision will depend on how your storage needs unfold in future years. If your storage needs are predictable, and storage will be needed for the entire useful life of the storage equipment, then a hosted storage solution would generally be best. However, if the future storage needs are unknown, cloud storage can reduce the risk of buying too much idle capacity or not enough spare capacity. Another consideration is the life cycle of your business. If you are in an early stage company where cash is a precious resource and loans (and possibly equipment leases) are often unavailable, cloud storage can help conserve cash for marketing and development while preserving the flexibility and agility needed in a start-up.
If they come, we will build it.
If your storage application needs are flexible to use both cloud storage and hosted storage, a hybrid solution may be best. As mentioned above, hosted storage is generally cheaper than cloud storage when storage needs are stable and predictable. However, we also discussed how the "build it and they will come" strategy has killed many a project burdened with cost of massive idle capacity. Also, the low cost/GB of hosted storage depends on the storage servers being used at or near capacity. One alternative is to turn the strategy on its head - "if they come, we will build it." Use low-risk flexible storage to accommodate new storage needs, and when a certain amount of storage need becomes evident, purchase enough hosted storage (servers near capacity) to off-load the storage from the cloud. This eliminates costly idle capacity, cloud storage can be provisioned almost instantly if needed, lower cost storage can be used once the need materializes, and cash is conserved until there is revenue to off-set the storage expense.
Cloud vs. Hosted Storage Excel Workbook [.xls] (231.00 kb)
b372e41c-570c-471d-b1c8-331f2d1fbd61|0|.0