4 ms·
The first key conclusion threw me a bit. Easy to misread as "Ignore UBER" rather than "Ignore UBER specs".
by vacri 11y ago
The first key conclusion threw me a bit. Easy to misread as "Ignore UBER" rather than "Ignore UBER specs".
- mvgoogler 11y agoThe article talks about ignoring UBER _specs_ the paper doesn't say that (at least that's not what I read). From the paper: We find that UBER (uncorrectable bit error rate), the standard metric to measure uncorrectable errors, is not very meaningful. We see no correlation between UEs and number of reads, so normalizing uncorrectable errors by the number of bits read will artificially inflate the reported error rate for drives with low read count." Basically uncorrectable errors are not correlated to the number of reads, so computing an UBER by dividing the number of uncorrectable errors by the number of reads is meaningless. Therefore UBER is a meaningless metric. They don't differentiate UBER from a manufacturer spec and UBER as measured in production.