Setup and Benchmarks of Several ESDI Hard Disks and Controller Cards

Test Setup

Most testing is done on a 486 ISA/EISA/VESA Local Bus platform (Gigabyte GA486-TA). Equipped with an Intel P24T Pentium upgrade cpu with 33MHz external and 83MHz internal clock speed, processor bandwidth shouldn't be a problem. Both RAM banks are filled with 16MB 30pin SIMMS, 128MB total. If both SIMM banks in a 386/486 systems are populated, ESDI data throughput is about 10% higher compared to single SIMM bank filled systems. Second level cache amounts to 512KB SRAM. ISA/EISA bus speed is adjusted to 8,3 MHz.

Floppy drive access and ESDI/SCSI bridge adapter testing are tasks of an Adaptec SCSI-II compatible host adapter (AHA2742-AT). During ISA/EISA ESDI controller tests, the SCSI host adapter BIOS of the AHA2742-AT is disabled by the AMI EISA configuration utility.

Generally the system BIOS hard disk is set to type 1. Some controller cards demand type 0 (not installed) for the first hard disk. ISA ESDI controllers are jumpered to use the first hard disk controller port and IRQ 14. If possible, the floppy controller port of the ESDI controller board is disabled, or set to the secondary port address respectively.

For the sake of commensurability, caching of the controller BIOS through system second level cache is usually turned off. Some cards do not tolerate BIOS caching, especially during low-level-format. As the only exception the OMTI 8620 BIOS is cached during performance tests. Its BIOS serves as a DOS device driver, which manages the hard disk access. BIOS caching pushes it's performance on a comparable level in relation to the other cards.

Micro channel (MCA) controller card tests are done on a 386DX-25 IBM 8580 with 64KB second level cache and 8MB planar RAM. Both RAM banks were populated with 4MB IBM RAM modules.

MS-DOS 6.22 serves as operating system since FreeDOS 1.0 refuses to start BIOS setup routines on several controller cards. Plain DOS is started from a floppy disk. The autoexec.bat file contains one line for the German keyboard driver only. There were no config.sys statements. No DOS memory management drivers, no disk caching software, or any other TSR programs are loaded.

For the low-level-format process the controller BIOS setup routine is used only in case if there is no DOS utility program from the controller manufacturer available. After the drive is low- level-formatted a method for addressing the drive's storage space has to be chosen.

If the drive has less than 528,482,304 byte storage space it is addressed using the so called '63 sector per track translation' mode according to the capabilities of the controller card.

If the drive has more than 528,482,304 byte storage space, in a first attempt the storage space up to 528,482,304 byte is addressed using the '63 sector per track translation mode'. In a second attempt the entire storage space is addressed using the so called 'head mapping mode' according to the capabilities of the controller card.

In any case, one primary startable DOS partition as large as possible is created. Then the system is rebooted, the new partition DOS- formatted, and the DOS system files installed on the hard disk. Again the system is rebooted from the hard disk, now ready for the benchmark test.

Benchmarking Software

The performance testing does SpinRite 5.0 from Gibson Research. It measures the data access time including rotational latency in milliseconds, the sector access velocity in megabyte per second, the burst data throughput between controller card and system bus in byte per second, and the sustained data throughput between drive media and system bus in byte per second.
Sector access velocity measures the rate of movement through unwanted data, which the drive's read/write heads overfly, while accessing a certain data sector. This value is dependent on the drive's access time, data and track density of the drive media and the drive's storage capacity. With the sector access velocity measurements it is possible to compare the performance of different drive-controller-combinations.
Additionally the available DOS formatted storage space is given in byte, because every controller manufacturer has a specific defect management and needs more or less storage space for partition management.

Since EISA controller need EISA drivers for full data throughput and optimized data access, Spinrite cannot measure the performance of these controller cards, while in EISA mode. Instead, EISA mode benchmark is done with CoreTest from Core International.

ESDI Data Throughput

The ESDI storage system consists of a disk drive and a separated controller board, which are connected through a 34pin drive control cable and a 20pin data cable, flat ribbon type both. Usually drive and controller card are from different manufacturers.

The hard disk operates with a specific data transfer speed, which the controller card has to match. In case of ESDI the data transfer speed is measured in MHz/s (megahertz per second) which is interchangeable with Mb/s. ESDI systems operate either at 10MHz/s, 15MHz/s, 20MHz/s, 22MHz/s, 23MHz, or 24MHz/s which correspond to 1.25MB/s (megabyte per second), 1.875MB/s, 2.5MB/s, 2.75MB/s, and 3MB/s. Controller cards support a range of data rates up to a certain data rate, e.g. a 15MHz/s controller cards is compatible with 10MHz/s and 15MHz/s drives, but incompatible with drives with a higher transfer speed. (Incompatible controllers and hard drives led to a lot of confusion, e.g. when a system was upgraded with a drive with higher capacity.)

The ESDI data transfer speed reflects the amount of unformatted bytes, which are exchanged on the interface between controller and drive. It has to be distinguished from the transfer speed for the formatted data, which is the data usable for the operating system and it has to be distinguished from the data transfer speed, the controller card is able to deliver on the system bus (e.g. ISA, EISA, or MCA).

The following two tables show basic features of the tested ESDI drives.

Table 11 measures the bitstream on the controller/disk drive interface. As you can see, the manufacturer's disk drive specification may differ from the 'mechanical' abilities of the drive. Some high end drives seem to use an own sector buffer. In that case the interface data transfer speed is higher than the read from/write to media data transfer speed. In other words; the minimum data transfer speed the controller has to accomplish may be higher than indicated by the amount of unformatted bytes per track.

In a process called low-level-format the controller formats the data on the hard disk in sector units. For the contemporary IBM-compatible PCs, a sector consists of 512 byte operating system accessible storage space and a variable amount of maintenance data. Since portions of the maintenance data are tied to particular time periods, the extent of maintenance data increases with the data transfer speed (...most of the time).

In an ESDI storage environment either the hard disk or the controller card decides about the actual sector length. Hard disks are equipped with jumper blocks or dip switches, which allow the user to set the drive in hard sector mode or soft sector mode. If the drive is hard sectored, the hard disk define the sector length in byte and thereby the amount of sectors per track. If the drive is in soft sector mode, the sector length is determined by the controller card.

Table 12 measures the formatted data throughput on the controller/disk drive interface. It is theoretical since the controller card reads one track always, so no head switch or head move occurs.
The fifth column shows the max. data throughput of the disk drive, which is usable for an operating system. Later we will use this value to measure the performance of the controller card and the system bus.

System Bus Data Throughput

Data throughput of ESDI storage depends on the disk drive performance, the controller card performance, and the system performance. In turn system performance depends in a fundamental way on the system bus used. The following table shows the bus architectures and their data bandwidth, which are used for performance tests.

This is no general overview of system bus performance. It gives an idea of the resources available to the ESDI controller card, if it would be able use the system bus for a whole second without being interrupted from other devices (e.g. graphic card). Since ESDI delivers a maximum payload of about 2,5MB/s, any of the above buses is able to transfer the data stream at a time.

The Results - 10 MHz Systems

First test is for the 10MHz drives. Tested are two hard disks, model 5655 from the Japanese manufacturer NEC, and model 94186-383 from the Control Data subsidiary Imprimis against 27 controller cards. I will start with the NEC 5655.

Some controller cards refuse to recognize the NEC 5655. To highlight the gaps between the cards, which operate the NEC 5655, measurements are indexed (Index 100=Average of 21 cards).

And now the same procedure for the Imprimis 94186-383 disk drive. Again the results are indexed in the next but one table (Index 100=Average of 24 cards).

The later Data Technology, Adaptec and Ultrastor cards show a remarkable good performance especially the DTC 6282. The first generation ESDI controller like WD1005 or OMTI 8620 cannot keep up with the later cards. Access times for both cards seem to be low on first sight, but due to the lack of translation methods they only address the first 1024 cylinder and not the 1224 and 1412 cylinder respectively the drives actually have. This is the cause for the huge storage space loss too. Furthermore the WD1005 cannot operate a 10 MHz hard disk with 1-1 interleave. To read a complete track, 2 revolutions of the drive's platters are necessary. As a consequence the sustained data throughput is low.
All controller cards, which use head mapping mode are slower than those, which address the storage space with the 63-sector-translation mode. More complex address translation seems to raise the data access time.
Perhaps more surprising is the relative low performance of the high-priced cache controller cards. The PM3011E/75 from Distributed Processing Technology (DPT), which was considered to be a top performer of it's day, operates a 10 MHz hard disks with an interleave of 3 to 1 and therefore achieves the worst sustained data transfer I've measured. A positive exception is CompuAdd's HardCache/ESDI which can keep up with the DTC controller cards.
The EISA controller cards provide very high burst transfer rates even without EISA software drivers loaded but, since they are all cached, access time is relative high and sector access velocity stays relative low.
The WD1007-MC1 achieves the best sustained data throughput measured, probably as a result of the shorter clock cycle of micro channel bus. Alas, micro channel architecture allows head mapping mode only, which slows down the ESDI performance significantly. Additionally the head mapping mode seems to turn off the track cache on the controller card. Since this relative small track cache (between 8KB and 64KB SRAM or VRAM) is responsible for the burst transfer throughput, ESDI storage in micro channel machines is relative slow.

The most fast controller card, the DTC 6282-24, will now benchmark the other available 10 MHz drives.

The IBM drive refused to work with the DTC 6282-24. Instead it is benchmarked with the two micro channel controller cards.

Now let's go on to benchmark results of the 15MHz systems.