Hardware Reference
In-Depth Information
applying the ECC information to correct it. Recoverable errors are those in which the data is read
incorrectly, but the drive uses the ECC information to correct the errors before passing the data on to
the system. An unrecoverable error rate of 1 in 10
12
bits basically means that one would expect to see
about one read error for every 125GB worth of data read from the disk. That was more than
acceptable when drive capacities were only 5MB to 100MB. You would have to read a 100MB
drive front to back 1,250 times before a read error might occur.
However, as drive capacities rose, the need for lower read error rates became necessary. The
original read error rate of 1 in 10
12
bits (one error every 125GB) would be unacceptable today,
because it would mean that reading a 1TB drive front to back just
once
would probably result in at
least eight unrecoverable errors! Fortunately, newer drives have lower error rates. Since the first PC
hard drives came out in the early 1980s, drives/controllers have been produced with primarily four
sets of unrecoverable error rates.
Table 9.5
shows the number of bits and terabytes read per
unrecoverable error for typical drives over the years.
Table 9.5. Bits/Terabytes Read per Error (Unrecoverable Read Error Rates) for Typical Drives
As you can see from the table, disk drive read error rates fell by a factor of 1,000 times in the first
5MB drives in the early 1980s through the first 1GB drives of the early 1990s, which is
understandable because drive capacities increased by a similar factor during that time. However,
since the early 1990s, read error rates have remained relatively stagnant, whereas drive capacity has
continued to increase by a factor of more than another 1,000 times. In other words, since the 1990s,
the reduction in error rate per bits read has not kept pace with the increases in drive capacity.
Most current consumer class drives advertise a 1 in 10
14
bits error rate, which means one error
should be expected for about every 12.5TB of data read. This is alarming when you consider that if
you were to read a 3TB drive front to back only five times, you are virtually guaranteed to have a
read error. Note that Enterprise/RAID class drives normally advertise a 1 in 10
15
bits error rate,
which is 10 times better than most consumer class drives, but still not what I would expect it to be
given the increases in drive capacity since that level was first reached.
Error rate reductions have not kept pace with capacity increases because of efficiency. To reduce the
read error rate, you must store more ECC bytes. In addition, as the areal density on the drive media
increases, small physical flaws or errors on the disk surface affect more and more bits, requiring
more ECC bytes and more complex calculations just to
maintain
a given error rate. A 12-byte ECC
was enough to achieve a 1 in 1014 error rate back in 1990; however, this grew to 24 bytes by 1999
and 40 bytes by 2004 just to maintain the same error rate. In other words, even without decreasing the
error rate further, the number of ECC bytes has grown from 12 bytes to 50 bytes, creating a significant
amount of storage overhead. By 2010, it was found that additional increases in ECC capability for
512-byte sectors would require the number of ECC bytes to increase to the point that it would negate
any further increases in density. In other words, a stalemate of sorts has been reached between density
and ECC capability.
