Quantum 10K II - Manual (Page 305)

Feature Descriptions
6-4 Quantum Atlas 10K II Ultra 160/m SCSI Hard Disk Drives
6.8 EMBEDDED SERVO SYSTEM
Embedded servo information is written in a spoke configuration on every track, on
every disk surface. The spokes (or headers) consist of quadrature analog patterns
and digital address data. The digital portions of the spoke data are read and used to
locate the desired track, spoke, and head number. The quadrature analog signal
portion is detected and used by a servo feedback control loop to precisely position the
head on the track center.
6.9 DATA INTEGRITY AND SECURITY
The disk drives use a combination of parity checking, error detection coding (EDC),
error correction coding (ECC), and checkpointing to protect stored data from media
errors, transfer or addressing errors, or errors introduced during block reallocation.
6.9.1 Media Error Protection
To ensure that data read is the same as data written, the drive computes and
appends an Error Correction Code (ECC) to each block of data stored. The drive uses
a 352-bit Reed Solomon code with a 4:1 interleave, which can correct up to 20 bytes
in each block.
The drive can also correct up to 2 bytes per interleave (up to 8 per block) in
hardware (“on-the-fly”), with no loss in throughput.
6.9.2 Transfer Error Protection
An end-to-end error detection code (EDC) protects data from any errors introduced
by internal buses, the disk controller chip, the data cache, or the SCSI interface.
An EDC is calculated and added to each data block as the data arrives from the
SCSI bus (after SCSI bus parity is checked). The EDC is stored with the data and
protected by the block ECC for added security. On reading or writing, the EDC is
checked as the data is transferred between buffer RAM and the media or the SCSI
bus.
6.9.3 Addressing Error Protection
Each data block on the media is identified and located by a servo spoke address. The
spoke address consists of a two-byte word. Each spoke has multiple copies of the
least significant bytes of the address. The disk hardware requires that a majority of
the copies agree and that the result agrees with the expected head, track, and spoke
number, before it will read or write the data.
To further protect against addressing errors, the logical address (LBA) of the data is
added to the EDC of each block. If data is written to the wrong block and
subsequently read, or read from the wrong block, the error will be flagged.
The hardware does not allow a blind read of a data block; the firmware must request
specific data blocks. Even if the head selection hardware malfunctions, it is not
possible for the drive to return data from the wrong head.