TLDR: While SSDs have a higher up-front cost, they are a large cost-saver in the long run in high-use scenarios such as data centers.
If you were building a large new datacenter, would you rather pay $0.10 per gigabyte for your storage, or $1.10? What if I told you that you should pay $1.10, and that it would save you almost 40% over 10 years? You would probably guess that I had flunked math (which luckily I only came close to doing in AB calculus), but there is a method behind the madness, and one that deserves a closer look.
Here’s the scenario: you are a part of an organization that is gathering historical scans from all around the world and will be archiving them for posterity. The data needs to be stored in a lossless format, and will collectively amount to 1.4 petabytes of data per year. Also, the images will need to be made available on-demand to lucky users across the Internet. How will you go about designing an infrastructure to handle these needs? This scenario is based on a real world implementation, and was given to us in an enterprise applications class with the instructions to create a feasible proposal for the project. While examining the various aspects of the project, we found that using solid state drives would be a huge cost saver in regards to total cost of ownership over the course of 10 years.
When storing large amounts of data, more than raw purchase price needs to be taken into account. Other important costs that must play in include factors such as:
- failure and replacement rates
- power costs
- cooling costs
- capacity (throughput/output) requirements
Typical enterprise-grade platter-based hard drives can cost as low as $0.10/MB to purchase, versus around $1.00/MB for a solid state drive1. Also, HDDs currently have much higher capacities than SSDs, with large SSDs typically maxing out at around 480 MB instead of 1 or 2 terabytes on HDDs. However, because solid state drives have no moving parts and run much cooler, they have lower failure rates. Furthermore, and more importantly when discussing costs, they draw drastically less power and require much less cooling than an array of hard drives2. Finally, when considering the need to serve read access to clients through the Internet, throughput becomes important. An SSD cluster can deliver on average 20-100x more throughput than can a comparable HDD cluster, even when properly RAIDed. Thus, the need for mirroring and splitting requests across drives drops significantly.
We plotted out the costs of an infrastructure using both traditional HDDs and new SSDs, considering the amount of drives that would need to be purchased at different times, the power and cooling costs, replacements, etc., and discovered that over the course of ten years, running a datacenter with SSDs would save an estimated $20,376,103.50, or 38%, when compared to the HDD option (HDD TCO: $52,535,121.04; SSD TCO: $32,159,017.54). While the first few years require a greater upfront investment in the actual purchase of drives, the savings in power, cooling, and replacements costs after year 5 begin to pay off substantially by the end of the product (see chart, full calculations available as a Google Spreadsheet here3). This result certainly surprised us, but it makes sense when you consider that adding space with hard drives is a very linear operation—the more drives the more heat and the more power. While solid state drives are pricier to purchase, their TCO is much lower when considered in mass quantities.
Many organizations are beginning to recognize this. Pure Storage is focusing on this angle, eBay recently deployed 100TB of solid state memory in their data center, and big data is really coming into its own. All of this just goes to demonstrate that we may be on the verge of a new and different data center, and that larger upfront costs may just pave the way for less expensive operational costs in the long run.
I certainly make no claims to being an infrastructure or hardware expert, or have experience in data center operations, but at least this was a good learning exercise for me. Remember, don’t discount options right away just because they appear to be more expensive at the outset!
- Denali, “SSD and HDD Economic Forecast: Analyst Jim Handy Speaks Out,” Jan 26 2010, http://www.denali.com/wordpress/index.php/dmr/2010/01/26/
- SSDs consume 15% the power of HDDs and have a 2 million hour MTBF lifespan. “Unified Storage for Dummies,” Oracle.
- Calculation assumptions are noted in the spreadsheet.
Share and Enjoy