I have some questions about the seek error rate SMART attribute. I have a Seagate 500Gb 7200.10 (about 3 years old) that has a high seek error rate.
I know that the raw data field has a different structure for different vendors and is thus not very meaningful. However, what concerns me is that the "current" and "worst" fields have also decreased a lot. I believe they initially start from 100, and have dropped to 47/42. The threshold is 30. For comparison, I have an older 120Gb Seagate (almost 2x power on hours) and its current/worst are 80/60.
HD Sentinel reports that the HDD health and performance are still 100% and the drive has more than 1000 days left. How can this be, seeing as the seek error rate has dropped from 100 to 42, is pretty close to the threshold and it is supposed to be a critical attribute. Other SMART attributes are still OK, no reallocated or pending sectors.
The seek error rate means that the drive is over- or under-shooting the correct track when it moves the heads, and it has to do another (small) re-seek to acquire the track before it can read or write the data. By itself, it's going to be a performance concern, but probably not that big a concern as far as data integrity goes.
As far as data integrity, pay more attention to the "Reallocation Count", and "Pending Sector Count" values as they indicate failures to read the data from the disk itself. If those numbers are high or increasing, then you should consider the drive to be unreliable and plan to get you data off it ASAP.
The normalised value of Seagate's Seek Error Rate is logarithmic. A value of 60 indicates that the drive has recorded 1 seek error in 1 million seeks, 70 indicates 1 error in 10 million, and 80 is 1 in 100 million. A value of 30 is 1 error in 1000, and 40 is 1 in 10,000.
A drive begins life with an SER of 100. After it records 1 million seeks the SER is recomputed to reflect the actual SER. Prior to that, the SER is considered to be statistically insignificant and remains at 100.
The raw value of the attribute records the total lifetime seeks in the lower 32 bits, and the total lifetime seek errors in the upper 16 bits.
The formula for SER is ...
normalised SER = -10 log (total seek errors / total seeks)