Updated May 24, 2022

What is DA Drive Analyzer?

DA Drive Analyzer ® is a software as a service developed by ULINK Technology that monitors various storage drive health indicators and utilizes a machine learning model (a type of artificial intelligence) to notify the user when their drive health is at risk of failure.

The model utilizes a deep neural network, which involves creating multiple layers of drive health predictors, with each predictor in each upper layer being derived from a pattern of predictors in the layer below it (the bottom layer consists of the original health indicators). The model has been trained and tested on data coming from a pool of data that consists of 1 million-plus drives and several thousand drive models, to give each predictor the proper weight to successfully predict drive failure. The weighting process is called “gradient descent” and involves minimizing the discrepancy between the model’s prediction and the observed data. During the development process, we have also systematically tuned non-data-driven hyperparameters to optimize the model’s prediction accuracy. Over time, our model will be tested with new client drives and updated as necessary to maintain a high degree of accuracy.

Advantages of a Machine Learning Approach

Previously, the health of storage drives was commonly determined by reading drive health indicators called Self-Monitoring Analysis and Reporting Technology (SMART) attributes off drives. SMART attributes serve as sometimes-critical indicators of drive health. And one way to utilize SMART attributes to predict drive failures was by looking out for SMART trips, which is when SMART attributes drop below manufacturer-specified thresholds. However, judging whether or not a drive will fail based on individual SMART trips does not catch many failing drives. SMART trips only manage to flag about 6% of failing drives (1). A machine learning approach can catch more failing drives than SMART trips.

Storage Drive Failure

As of 2022, individual drives from some companies still have an annual failure rate of greater than 2% (2). If you are a business or consumer whose data is stored on one of these drives, then you run the risk of losing valuable work. Relying on SMART trips will cause you to miss many drive failures. Thresholding SMART attributes in other ways, while doable, requires you to manually balance catching true drive failures with keeping false alarms low -something that requires a lot of time and a lot of drives to successfully carry out. Conveniently, DA Drive Analyzer’s ® machine learning algorithm has been trained to maximize the number of drive failures it can catch while keeping the false alarm rate at an acceptable level.

The ULINK Advantage

ULINK Technology is a company that has performed storage testing for companies such as Intel, Seagate, and Western Digital. By leveraging our storage testing experience, for HDD’s and SSD’s, we are able to collect a variety of drive health information beyond SMART attributes. This gives us a wide pool of drive health predictors to work with and allows our machine learning algorithm to utilize more than just SMART attributes. As a result, can offer machine learning as a practical solution for predicting drive failures, available for a variety of drive models.

With data becoming more important than ever to businesses and consumers, can people afford to be uncertain about the fate of their storage drives? ULINK’s Drive Analyzer ®, driven by machine learning technology, is our answer.

References:
1. Based on analyses of data from QNAP Systems by ULINK Technology.
2. https://www.backblaze.com/blog/2018-hard-drive-failure-rates/ .