Hard drive reliability has long been a critical concern for data centers and individual users alike. A new peer-reviewed study published in the IEEE journal has shed light on this important topic by analyzing 12 years of hard drive performance data from Backblaze. The research, conducted by Christoph Siemroth from the University of Essex and Yeomyung Park from Sungkyunkwan University examined 443,156 hard drives representing more than 1.66 million drive-years of operation.
The study provides valuable insights into how different hard drive manufacturers perform under similar operating conditions, offering guidance for anyone making storage decisions. The findings reveal significant variations in reliability across major brands, with some surprising results about which drives are most dependable.
HGST emerges as the most reliable brand despite being discontinued
The research found that HGST drives had the lowest failure rates when compared to other brands under matched conditions of age, capacity, form factor, and operating temperature. Specifically, HGST drives failed at only 41% of Seagate‘s rate, making them the most reliable option in the study. Western Digital (WD) came in at 52% of Seagate’s failure rate, while Toshiba had the highest failure rate at 107% of Seagate’s. Interestingly, HGST drives are no longer available for purchase new, as WD acquired the brand in 2012 and subsequently wound it down.
The study highlights an important limitation: the comparison is based on enterprise drives used in Backblaze’s data centers, which may not directly translate to consumer models. The researchers note that Backblaze’s quarterly reports often compare drives from different eras, which can skew results. For example, HGST led new installations in 2014, Seagate dominated from 2015 to 2026, Toshiba topped 2026, and WD led in 2026. This means raw annualized figures often pit aging drives against newer ones.
Toshiba shows dramatic reliability decline after 60 months
One of the most striking findings was Toshiba’s dramatic increase in failure rates after 60 months of service. The study found that Toshiba’s monthly failure rate more than quadrupled at this point, reaching five to six failures per 1,000 drives per month between 50 and 100 months. In contrast, the other brands generally maintained failure rates at or below 2.5 per 1,000 drives per month during this period. Seagate’s failure rate started highest among young drives but remained relatively steady with age.
Backblaze’s own analysis had previously shown that most drive failures occur before three years of service. The new study confirms this pattern while adding important nuance about long-term reliability differences between brands. The researchers also noted that 146,943 drives, about 31% of the sample, exited the dataset without failing, primarily due to capacity upgrades.
Temperature and capacity affect reliability in predictable ways
The study identified several factors that consistently affected hard drive reliability. Each additional one-degree Celsius increase in average drive temperature raised failure rates by 2.1%, compounding to roughly 23% across a 10-degree increase. Conversely, each extra terabyte of capacity reduced failure rates by about 3.4%. The researchers attribute this improvement to advancements like helium-filled enclosures and newer head-fabrication processes in higher-capacity designs.
The dataset’s limitations are important to consider. Only 0.52% of sampled drives ran longer than a decade, so the estimates primarily reflect short- to medium-term reliability. Additionally, Backblaze doesn’t record comparable workload data across manufacturers, which could potentially affect the results. The authors suggest that WD’s ownership of HGST might explain why these two brands showed the closest failure rates, possibly due to shared technology.
As of, Backblaze’s most recent annual report put its fleet-wide annualized failure rate at 1.36%. This comprehensive study provides valuable data for anyone evaluating hard drive options, though the enterprise-focused nature of the dataset means consumer users should consider these findings in context.
