Understand what your system needs first
Just this past week I found myself discussing the merits of yet another IoT Platform as a Service (PaaS) offering with a sales representative. It did not take too long for the conversation to take the obvious turn toward the number of connected devices the platform could support and how well it scales. The typical discussion starts out with the vendor saying that a single instance can support about 50,000 connected devices, followed by the typical speech on how the platform scales horizontally to support tens of millions of connected devices and billions of messages. I have heard the speech so many times and each time the potential vendor is making promises without knowing anything about your products or the needs of your solution. What is perhaps the most interesting is that most companies on the market for an IOT PAAS are not well enough informed about the needs or characteristics of their own products. If your IoT strategy is going to be successful, you need to do your homework and clearly understand your sales projections, the demand each connected unit is going to put on the system and finally, how much data you can expect to send to and from the cloud before you select or build your IoT Platform. If you do, you’ll find that if your IoT Strategy is going to be successful you need a strong edge strategy and solution.
All companies go through the process of forecasting their sales numbers for one or more likely 2 to 3 years. However, as I talk with engineers throughout the industry I realized that the technical teams responsible for connecting the companies’ products to the IoT Platform in the cloud seldom have the accurate forecasts to inform their technical decisions. Some may only know the manufacturing numbers or perhaps the best case sales expectations. As an example, I remember in a previous role, I was told we’re expecting to manufacture 100,000 devices that one year, but there was no expectation set for the number of items that would sell, and no prediction for how many would actually be connected to the internet. I suspect as you read this; you’ll agree that manufacturing numbers alone are not enough to accurately predict the load on the system.
Even if the technical team had an accurate sales projection that only would not be enough to answer the question, “can this system handle the load?” There is a critical piece of information is missing from the equation, the demand characteristics. It’s the demand characteristics such as the number of sensors or variables on the devices that can change and even more important how quickly they can be expected to change. Let’s imagine for a moment our connected device consists of 10 distinct variables with each variable expected to change once per minute. The demand characteristics of this device would be 10 UPM (updates per minute) for 100,000 connected products results in 1,000,000 UPM, roughly equivalent to about 16,600 database writes per second.
If your IoT strategy is going to be successful it’s important that you have good sales and connectivity forecast for at least 3 years. But that is not enough; you need to have a clear understanding of the demand characteristics each connected device is going to put on the system in the cloud. These two values will enable you to predict how many updates per minutes should you expect on average per device. Finally don’t forget to consider the amount of data you’re going to be sending to the cloud on a regular interval. Once you’ve completed your analysis and understand the load on the cloud, you’ll be informed enough to consider and formulate a strong edge solution that will ensure the overall solution meets your needs at a reasonable cost.