Quantum ProDrive SCSI

The following document should be read in order to understand some issues which may arise during formatting and bench marking the Quantum drive as installed in the Qi 900.

The Quantum ProDrive is a 3.5" form factor SCSI Winchester drive that adheres to the ANSI X3T9.2/82.2 Revision 17B SCSI specification at conformance level 2 with Command Set Revision 4B. It is installed as original equipment in Qi 900 series machine. The drive is "hard" sectored to provide physically 512 bytes per sector (plus 6 bytes for Error Correcting Code (ECC)) with every 6th track containing a reserved sector for defect management.

ERROR CORRECTION

Defect Definition (General Note)

Any discrepancy between recorded data and recovered data at the sector level is a data error. In high density digital recording systems, it is necessary to provide an error detection and correction scheme to enhance the operational performance and increase the reliability of the system.

Data errors can be either "soft" (not readily repeatable) or "hard" (repeatable with high probability). Soft errors are generally related to the signal-to-noise ratio of the system. They represent marginal conditions of the media, heads, read/write circuitry, and the controller data synchronisation circuit. Hard errors are most often due to defects, pits, scratches, or thin spots in the media surface. These defective media areas can be detected and skipped (not used for storage).

Defect Management (General Note)

Disk drives are scanned by their manufacturer for defective sectors on the disks. Those defective sectors are deallocated prior to shipment. A factory Defect Map, duplicated on tracks on the drive, allocated for this purpose, will contain the location of all defective sectors found in the drive.

One sector is allocated per defect zone as a replacement sector. Defects found at the manufacturer are skipped using an "In-Line" sparing method, which replaces each sector with the next sector on the track. In the rare instance where more than one defect is found within the same zone, the additional defects will be mapped into the nearest adjacent zone with a spare sector available.

Errors reported during field use (disk or data errors), are detected by SCSI aware operating systems (DOS does not fall into this category) and stored in a temporary list, these areas are then avoided for data storage. To map these defective areas out of normal use a low level format programme will need to be run, to in line sector spare the drive.

Low Level Format On the Quantum Drive (SFORM)

In general, when SFORM is run on a SCSI drive, the manufacturers defective media list is read and any defective areas skipped using the "In-Line" sparing method detailed above. The drive is then scanned to verify sector headers, and any additional defective areas mapped as bad.

On the Quantum drives however, The only operation performed is "In-Line" sparing after reading the manufacturers defect list. This typically only takes a few seconds, and results in the drive appearing to have all its data still intact. As no read write test is performed, any additional grown defects will not be mapped out. To correct this HDSETUP.EXE is to be amended to include a format unit command for Quantum drives which will do a media scan in an attempt to detect grown defects.

Quantum ProDrives also have the facility to detect and add grown defects to an internal list. This area will then automatically be re-mapped to the first available free sector, including data if possible. This operation will not in line sector spare the drive but will make grown defects almost transparent to the user. As default this option is disabled, but it is intended to enable it during the enhancement of HDSETUP.

Quantum's Disk Cache (DisCache)

Quantum's Disk Cache anticipates host-system requests for data and stores that data for faster access. When the host requests a particular segment of data, the caching feature uses a pre-fetch strategy to "look-ahead" and automatically store the subsequent data from the disk in RAM. If the host requests this subsequent data, the RAM is accessed rather than the disk.

Since typically 50% or more of all disk requests are sequential, there is a high probability that the subsequent data requested will be in the cache. This cache data can be retrieved in microseconds rather than milliseconds. As a result, Quantum's Disk Cache can provide substantial time saving during half or more of all disk requests.

This drive based cache works by continuing to fill its cache memory with adjacent data after transferring data requests to the host. Unlike a non-caching controller, Quantum's SCSI disk controller continues a read operation after the requested data has been transferred to the host system. This read operation terminates after a programmed amount of subsequent data has been read into the cache segment (Maximum Pre-fetch). During write operations, this cache employs a write-through design, in which data written to the drive is written immediately to the disk. This design ensures data integrity in case of power loss.

The pre-fetch cache may be configured to offer an optimum performance for any given operating environment. However all drives fitted into Qi 900 machines will be configured to the default option of 4 segments of 16KB for reading purposes, 1 sector minimum pre-fetch after each read command and 16 sector (8KB) maximum pre-fetch after each read. This will provide optimum performance over a wide range of operating environments.

PERFORMANCE

DOS Bench Mark Testing

In light of the operation and configuration of Quantum's cache, as mentioned above, the results produced by some industry standard bench mark tests, may not reflect an accurate record of the true performance characteristics of this range of drive. In short, by re-programming the drive cache significantly different results may be obtained from the same bench mark test.

The operation of some of the more common bench mark tests is examined below.

Core Vr 2.8

This test attempts to assess a drives performance by measuring its sequential transfer rate as well as its Average and Single track seek timings. Unfortunately the Adaptec AHA1640 SCSI Card emulates an INT 13h DOS call by presenting the systems BIOS with a drive which appears to have 64 Heads and 32 Sectors per track at 512 bytes per sector, i.e. the drive always appears to have 1MB Cylinder Capacities. A single track seek timing measurement undertaken through Int 13h will therefore end up seeking between these imaginary 1MB Cylinders which at best is a 10 track physical seek on this Quantum Drive. Such measurements are therefore meaningless.

The same argument applies to Average seek timing measurements, although because of the Random nature of the seeking, the results are not always totally inaccurate. There is though, a bias in favour of higher capacity drives with higher cylinder capacities even if the quoted access time for these drives is identical. To assess the random performance of SCSI drives in general, under the DOS operating system it is best to utilise randomised Int 25h single sector read calls.

Power Meter Vr 1.3

Power Meter attempts to emulate a typical DOS user environment under various application packages. It effectively assumes that the packages themselves are sequentially laid out on the disk as achieved under a normal installation process. For this test the Quantum drive would give best performance if its cache is configured to one segment of 64KB.

----------oOo----------