Data Lakes For Dummies. Alan R. Simon

Чтение книги онлайн.

      FIGURE 1-3: Different types of data in your data lake.

In a database, you define columns (basically, fields) for each of your pieces of structured data, and each column is rigidly and precisely defined with the following:

      Unstructured data: A picture may be worth ten million words

      Unstructured data is, by definition, data that lacks a formally defined structure. Images (such as JPEGs), audio (such as MP3s), and videos (such as MP4s or MOVs) are common forms of unstructured data.

      Semi-structured data: Stuck in the middle of the lake

      Semi-structured data sort of falls in between structured and unstructured data. Examples include a blog post, a social media post, text messages, an email message, or a message from Slack or Microsoft Teams. Leaving aside any embedded or attached images or videos for a moment, all these examples consist of a long string of letters, numbers, and special characters. However, there’s no particular structure assigned to most of these text strings other than perhaps a couple of lines of heading information. The body of an email may be very short — only a line or two — while another email can go on for many long paragraphs.

      In your data lake, you need to have all these types of data sitting side by side. Why? Because you’ll be running analytics against the data lake that may need more than one form of data. For example, you receive and then analyze a detailed report of sales by department in a large department store during the past month.

      Then, after noticing a few anomalies in the sales numbers, you pull up in-store surveillance video to analyze traffic versus sales to better understand how many customers may be looking at merchandise but deciding not to make a purchase. You can even combine structured data from scanners with your unstructured video data as part of your analysis.

      If you had to go to different data storage environments for your sales results (structured data) and then the video surveillance (unstructured data), your overall analysis is dramatically slowed down, especially if you need to integrate and cross-reference different types of data. With a data lake, all this data is sitting side by side, ready to be delivered for analysis and decision-making.

      Different water, different data

      A common misconception is that you store “all your data” in your data lake. Actually, you store all or most of your analytic data in a data lake. Analytic data is, as you may suspect from the name, data that you’re using for analytics. In contrast, you use operational data to run your business.

What’s the difference? From one perspective, operational and analytic data are one and the same. Suppose you work for a large retailer. A customer comes into one of your stores and makes some purchases. Another customer goes onto your company’s website and buys some items there. The records of those sales — which customers made the purchases, which products they bought, how many of each product, the dates of the sales, whether the sales were online or in a store, and so on — are all stored away as official records of those transactions, which are necessary for running your company’s operations.

      But you also want to analyze that data, right? You want to understand which products are selling the best and where. You want to understand which customers are spending the most. You have dozens or even hundreds of questions you want to ask about your customers and their purchasing activity.