Backblaze’s latest report, from one of the most respected names in the storage industry, continues to shed light on SSD failures through its detailed analyses. In this first-half 2023 report, the company examined the performance and reliability of more than 3,000 SSDs used in a professional, high-intensity environment.
For several years now, Backblaze has offered valuable insights into hard drive durability through statistics gathered in its data centers. The company has recently expanded its scope of analysis to include SSDs as well. This data is particularly relevant, as it is based on hundreds of units running continuously, 24 hours a day, 7 days a week.
Although SSDs are primarily used as boot drives for servers, their roles extend further. As Backblaze points out, they also handle log and temporary file storage, reflecting the daily activity of the storage server. For perspective, at the end of 2022 the company was operating 2,906 SSDs. That number grew to 3,036 by the end of March 2023 and 3,144 by the end of the second quarter.
This report promises in-depth insights into SSD performance in a demanding environment, while highlighting important nuances to consider when evaluating failure rates.
Half-Year SSD Failure Results in Backblaze’s Data Centers
As of June 30, 2023, Backblaze’s storage servers held 3,144 SSDs, a 20% increase compared to the previous 2022 report. 238 SSDs were added to the fleet used for failure analysis. The Crucial model (CT250MX500SSD1), with 110 new drives, is the most represented, followed by 62 new WDC drives (model: WD Blue SA510 2.5) and 44 Seagate drives (model: ZA250NM1000).
It is essential to note, as Backblaze rightly points out, that interpreting statistics — especially those relating to drive reliability — requires careful consideration of sample size and confidence interval. Drawing hasty conclusions without accounting for these factors could lead to inaccurate or misleading interpretations.
In such cases, there may simply not be enough data to produce meaningful results. That is why, for each drive type, Backblaze prefers to have at least 100 drives and 10,000 drive days within a quarter before considering the estimated AFR “reliable”. For accurate results on a specific drive model, it is important to take into account the number of drives and drive days before analyzing the calculated AFR (Annualized Failure Rate).
SSD Failure Rates Over Time
Quarterly data can be volatile due to factors such as drive age or the random occurrence of failures, which can push the AFR up or down. As shown in the chart above, the AFR fluctuates between 0.36% and 1.72%. Quarterly rates therefore do not always provide valuable information. Sometimes certain drive models turn out to be incompatible with the system environment, leading to their removal from circulation and reducing or eliminating their impact on the overall analysis.
Although these drives don’t play as critical a role as data drives in maintaining system robustness, it is essential to keep boot drives active in order to collect the information they log daily. In cases of incompatibility, it becomes necessary to adjust for — or even eliminate — the influence of the faulty drive and consider the AFR critically.
Backblaze’s Internal Use of the Data
As shown in the SSD and HDD Drive Stats reports, Backblaze generates quarterly, annual, and lifetime charts from the data it collects. Each chart usually represents a specific drive model, but in some cases several SSD models are grouped together into a single chart.
For example, the Micron SSD results in the chart above show the need for deeper analysis of the underlying data. By continuously collecting, storing, and analyzing Drive Stats data, Backblaze can proactively maintain its durability and availability targets. Without this data, the company could be inclined to overwork its systems, remaining unaware of the random nature of SSD failures that directly affect data center performance.
A First Look at Other SSD Statistics
In addition to quarterly and lifetime failure rates, the quarterly Hard Drive Stats reports have examined other indicators over the years. The chart below offers an initial look at two of these indicators : the average age at which SSDs fail, and how closely their failures follow the bathtub curve.
The “bathtub curve” is a concept used in reliability engineering to describe the failure rate of a product or system over time. This curve is divided into three distinct phases:
- Early failure phase (or burn-in period): At the start of a product’s life, there can be a high failure rate due to manufacturing defects or other initial issues. These failures are often identified and corrected quickly.
- Constant failure phase (or useful life): After the burn-in phase, the product enters a period where the failure rate is relatively constant and low. This is generally when the product is most reliable.
- Wear-out failure phase (or end of life): As the product ages, its components begin to wear out and the failure rate rises again.
The value of this curve lies in helping engineers and designers understand and predict a product’s or system’s failure behavior throughout its life, and identify opportunities to improve reliability.
Although these datasets are limited in size, they still offer valuable insights. According to Backblaze, while the current curve (blue line) showing SSD failures for each quarter is somewhat “irregular”, the trend line (red) does have “the shape of a bathtub curve”.
The trend line matches about 70% of the actual data, so Backblaze states it is not fully confident at this stage, but given the limited amount of data available, it appears that SSD failure occurrences are on track to conform to the well-established bathtub curve.
You can find all the data from this study on the resource page made available by Backblaze.