A FAST '16 paper titled "Flash Reliability in Production: The Expected and the Unexpected," by Professor Bianca Schroeder of the University of Toronto, and Raghav Lagisetty and Arif Merchant of Google, studied the reliability data from millions of SSD drive-days over a period of 6 years, to come up with some very interesting conclusions on SSD reliability. One of the study's biggest finding is that high-end (read: enterprise) SSDs with single-level cell (SLC) NAND flash memory are no more reliable than cheaper multi-level cell (MLC) drives. Besides millions of drive-hours, the group also studied 10 different models of enterprise and consumer SSDs, from three different memory types - MLC NAND, SLC NAND, and eMLC NAND.
The study also shows that RBER (raw bit error rate) is a more reliable measure of reliability than UBER (uncorrectable bit error rate) mentioned in drive specs or datasheets. RBER increases slower than expected from wearout, and isn't correlated with UBER. However, the measured/real-world UBER is higher for SSDs than HDDs. This means what while SSDs are less likely to "fail" than HDDs, they're more likely to lose portions of their data. Keep your SSDs regularly imaged. Age, rather than usage, affects reliability of SSDs. A disturbing 30-80% of SSDs in the study developed at least one bad-block, and 2-7% of the SSDs developed at least one bad chip, within the first 4 years of deployment.
http://ift.tt/1RyqtyT
Комментариев нет:
Отправить комментарий