4/17/91 The following model was originated by Doug Pickford. Dal Allan and I accepted an action item to massage it into a form which we thought was appropriate as a starting point for discussion on its merits. Dal produced a cut which is the file PICKFORD.DAL. I revised his version and produced this file PICKDAL.GEM. This file has my additions underlined. Deletions are not shown at all but were generally located in the vicinity of an addition. As I have noted in the past I am generally leery of an effort to standardize DCS because it limits technology progress and is very susceptible to being misunderstood and/or misapplied. I have also stated that if in spite of these dangers a viable standard emerges from these dangerous circumstances, I would be willing to support it. The following is offered for discussion but I certainly have not reached agreement with it. Gene Milligan X3T9.2 Participant 8.14 Direct Access Devices, Physical Layer Model The Logical level interface provided by SCSI was not intended to deal directly with the physical characteristics of Direct Access Devices which are comprised of multiple elements e.g. mechanical components, Analog Circuitry and Digital Circuitry. Disk drives operating as SCSI targets present consistent data to initiators, despite the wide variations in technology and differences in algorithms used for data recording and recovery e.g. Offsetting the Read/Write Head, Shifting the Data Window, Application of an ECC and/or Use of Retries. The purpose of the Diagnostic Command Set (DCS) extensions is to control these algorithms. The results may be applicable to an evaluation of the design and the manufacturing process for the physical device. However interpretation of the results requires substantial insight into technology which is undergoing continuous evolution and in some cases revolution. The descriptive material on the characteristics of disk devices is not inclusive of all the technology used. It is intended to provide a perspective of the kinds of capabilities which the Diagnostic Extensions can impact. 8.14.1 Physical Geometry The physical device consists of some number of recording disks which are used to store information. The information includes, but is not limited to, user data, sector identification, system operation parameters and actuator positioning data. The smallest addressable entity by the normal I/O is called a logical block. The smallest addressable entity by diagnostics are called sectors, which are sequentially organized in concentric circles called tracks. A sector may contain multiple logical blocks and/or a logical block may require multiple sectors. The beginning of a track is indicated by a signal traditionally called INDEX. The DCS extensions refer to the track start marker as INDEX. A collection of tracks which are accessible on all disks without substantial actuator arm movement is called a cylinder. Some devices read and write sectors by means of magnetic heads attached to a linear or rotary actuator. The motor driven (stepper or voice coil) actuator is positioned over the correct cylinder and a head (located over one track) is selected. If the servo information is on a track basis such as embedded the selection must be completed before positioning is completed. The requested sector(s), and other sectors, as they come under the Read/Write head are passed through the read channel to the controller for processing. The recording surfaces are not required to be perfect in order to make use of the media. A limited number of sector locations may be identified as containing a defect during the production process, or during actual usage of a drive. SCSI drives use a variety of Defect Management algorithms to mask the presence of the sectors with defects from the user. SCSI transfers only the User Area of disk cylinders. There are more cylinders over the entire stroke (length the actuator can travel between the physical ends of its range of movement). The ends of the stroke are protected by some means referred to as crash stops to catch "runaway" actuators. The non-User data cylinders may play a large role in the management of the disk device and may provide additional elements to evaluate the drive characteristics. The time it takes to complete a rotation is dependent on the rpm (revolutions per minute), which can vary e.g. the rotation time for 3600 rpm is 16.67 milliseconds and for 5400 rpm it is 11.11 milliseconds. Latency is the time it takes between the time a head settles and the drive is ready to transfer the requested data. The average latency is half the rotation time. The MODE SENSE/MODE SELECT commands provide a means for the user to be aware of, and to some degree, control the personality of the physical device e.g. the Cylinder and Track/Sector Map Mode pages provide a means for a disk device to report the detailed physical characteristics of drive geometry, 8.14.2 Servo System for Motion Control The actuator is controlled via a motor to position the head over the correct data track. Positioning information recorded on the drive is a means to inform the servo system of the current location. Should this information become corrupted, positioning to specific tracks may become degraded. Different types of servo systems include: * Open Loop systems: Controlled by only a Stepper Motor with no feed back information during a seek operation. * Closed Loop systems: Controlled by either a Stepper Motor or a Voice Coil Motor with positioning information located on a dedicated surface or embedded within the data surface. Different types of embedded servo include: - sector servo systems, whereby the positioning information is located in fixed locations relative to the sector data - wedge servo systems, whereby the positioning information is located in fixed locations relative to a consistent position such as Index, and may cause additional gaps in the data fields. There are many elements involved in a servo system: - seeking: moving a single and/or multiple cylinders - switching: the changing of Read/Write heads without changing cylinders - settling: the time it takes after physical motion for the head to be declared "on track" after seek completion - algorithms: various means to optimize seek motion or track following Possible candidates to evaluate seek motion include: * Random Seek tests: Stresses seeking over complex cylinder patterns and its impact on the servo system. This can require extensive run time to achieve a representative sample. * Sequential Seek tests: Stresses the single cylinder seek portion of the servo system. * Write/Read tests: Stresses the settling times of the servo system by using the above seek tests in combination with reading and writing of data. Checking the validity of recorded data during such a test may be a means to evaluate Read Channel performance, in addition to the Servo system. 8.14.3 Data Recording and Recovery The recording and recovering of magnetic disk information is done via a magnetic head which can either induce a magnetic field (writing), or detect a magnetic field (reading). While the Host System, Initiator and Target Controller view data in terms of bytes (00h through FFh), the Read/Write Channel sees magnetic field changes and/or magnetic field strength. Data encoding schemes are employed to optimize the efficiency of the read/write channel. Amongst the schemes used are methods generally known by the name of Run Length Limited (RLL) encoding, of which there are numerous types referred to by some of their code parameters such as 1:7 and 2:7. It is possible to use the DCS to infer some of the Head-Media characteristics by the Diagnostic Erase Track and Diagnostic Read Track operators of the SEND DIAGNOSTIC commands. Both reading and writing are affected by the Motion Control System's ability to maintain adequate centering of the head over the requested track. The process of writing is not usually checked at the instant of recording. It may be accomplished via a subsequent read, usually a Verify in SCSI, at the same location. The ability to reliably recover information on the data disks is influenced by many factors, such as media defects, read channel quality, system resonances; characteristic frequencies of the system, and noise. The Read process is also affected by the data window, which is the time slot when the Read Channel is expecting to find magnetic information to be interpreted as data. If the Read Channel attempts to read too early or too late, data integrity may be compromised. The window "moves" if the writing process is also influenced by window shifts (intentional or otherwise). Window shifts may optionally be controlled by the R/W Error Recovery Mode (Select) Page. 8.14.4 Data Integrity The most important factor in any data storage and retrieval device is the ability to accurately maintain data. Several means have been developed by drive manufacturers to compensate for less than perfect heads, media and positioning systems. Some of the tools that can be controlled by the device as well as accessed via the DCS command set include: * Data Strobe Offsets (Window Shifts) * Head Offsets * Error Correction Algorithms * Retries * Detection Parameters * Adaptive Algorithms During the certification pass of a Format Unit command, a disk drive may employ several or all of the above to ensure the media can be written and read under all conditions. 8.14.5 Drive Qualification Some purchasers of disk drives conduct a lengthy and thorough investigation of devices which they are considering for purchase, in a process generally referred to as qualification. Other purchasers shorten this process by sharing in the qualification conducted by the drive manufacturer. This enables them to reach the market sooner. During qualification, the device is scrutinized in many ways. A variety of tools such as those defined by DCS provide a means to assess: * Probability of Errors * Quality of the Servo System * Flexibility of the Read Channel * Design Margin of the Product * Conformance to the Specifications * Adequacy of the Defect Management One method used during qualification relies on reading and writing of data in a controlled environment. The Diagnostic Read Track Interleave - SEND DIAGNOSTIC and Diagnostic Write Track - SEND DIAGNOSTIC commands permit limited area media formats at specific locations on the media e.g. Outer Diameter and Inner Diameter. The purchaser may wish to erase all information from a limited area of the disk and determine the noise level or control parameters such as Window Shift, Peak detection thresholds and Head offsets to confirm the blocks resident in the P-list. The Translate Address Page - SEND DIAGNOSTIC command can be used to convert newly assigned physical addresses created with Diagnostic Write Track - SEND DIAGNOSTIC to Logical Block Addresses. The Diagnostic Erase Track - SEND DIAGNOSTIC and Diagnostic Read Track - SEND DIAGNOSTIC commands permit a unique assessment of the raw head-media-read channel combination. To control some of the device algorithms, Mode parameters are provided to manipulate: - Head Offsets (R/W Error Recovery), - Data Strobe Offsets, also referred to as Window Shift (R/W Error Recovery) - Data Detection Thresholds (R/W Control) - Read and Write Retries (R/W Error Recovery) - the use and extent of ECC (R/W Error Recovery)