VOICE Home Page: http://www.os2voice.org

March 2002 [Newsletter Index] [Previous Page] [Next Page] [Feature Index] editor@os2voice.org

The SCSI Workshop

[Eric Baerwaldt has composed a CD accompanying his SCSI workshop, which contains technical documents (amongst other things of more than 150 harddisks), drivers etc. This CD is available directly from the author for EUR 13,00 for german customers / EUR 15,00 for foreign customers incl. postal charges and shipment. Interested party's please transfer this amount to his bank account 5711 81 841, Sparkasse Nürnberg, BLZ 760 502 10 and send an e-mail with their postal address to EricBaerwaldt@web.de. The CD will be shipped instantly after the amount has been received. - The Ed..]

Any OS/2 user or eComStation owner planning to buy a new PC or to upgrade an older machine these days will inevitably have to deal with the following question: Should the new PC or the machine to upgrade be equipped with a SCSI subsystem (SCSI stands for Small Computer System Interface) to control its mass storage devices (harddisks, CD-ROM drives, streamers etc.) or should it host a conventional EIDE controller? Of course the very next question in this context will be:
Which advantages and disadvantages do these different principles offer?

To get right to the point: Except for its advantage of being low cost because of its mass production and ease of configuration because of its rather primitive interface design, EIDE doesn't offer any advantages when compared to SCSI. The theoretical cost advantage of EIDE systems instantly falls short when we plan an upgrade of an older machine. Such an upgrade doesn't make much sense for EIDE systems because attaching a new and faster harddisk doesn't automatically introduce the desired gain of speed. With SCSI systems this is completely different: a new controller with a fast SCSI harddisk can make even some older systems run with incredible performance.

On the other hand, as with every sophisticated technology, the configuration of a SCSI system is more complicated, so we'll deal with the SCSI technology in this issue (and those to come).

I. What is SCSI anyway?

History

In the world of IBM-PCs and compatibles on the other hand there were technological concepts from the stone ages at that time: even until the late eighties these machines had MFM controllers driving ST 506/412 hard disks, whereas other peripheral devices like scanners or tape drives were usually shipped with proprietary controllers which of course needed at least one unoccupied slot, and also took some of the scarce system resources like interrupts and dma channels in those machines, most of which were based on the ISA bus. With the introduction of the PS/2 series in 1987 however IBM decided to switch to more advanced technological concepts: at first the smaller PS/2 machines had proprietary IDE controllers and appropriate harddisks while the larger ones mostly sported ESDI subsystems, a successor to the ST506/412 technology offering enhanced performance. IBM even managed to put 2.5" ESDI harddisks into the first notebooks of their PS/2-Note series in the early 90's. However it turned out quite quickly that this mix of different and mostly incompatible interface designs for all of those peripheral devices as well as the technical restrictions of the ESDI standard asked for a more modern concept. (see ¹) So, when IBM corporation introduced their 3rd generation of PS/2 systems in 1992, they began to build SCSI subsystems into the premium line of their PS/2 machines, partly as onboard solutions and partly using separate interface cards. In 1994 the so-called fast SCSI standard aka SCSI-2 passed the ANSI committee, and since then SCSI-3 and several other enhancements based on it have lead to enormous technical innovation.

These rapid technical improvements however have provided us with an almost inscrutable number of interface cards, plugs, configuration schemes and mass storage devices, so it seems necessary to cast some light on this mysterious subject.

Technical basics

All communication on that bus is carried out in three stages (from a somewhat simplified view):

• In phase one, the command phase, the controller and the device negotiate the data transfer mode and then the device receives its command. The device (usually a hard disk or a CD-reader) can then decide to either lock the bus or disconnect itself from it as long as it takes to execute the command thus enabling the bus to deal with other communication tasks, and then reconnect/reselect after it has carried out the command received from the controller.
  
• In phase two which is the data transfer phase, controller and device exchange the very data which the controller then passes to the cpu.
  
• In the third and also last phase, the status phase, controller and device exchange status information thus implementing a powerful error correction feature making it possible to, for example, repeat the data transfer if there might have been an error in the bus system.
  

The processes described above, which are being executed without the help of the cpu, are referred to as bus mastering. So bus master capable controllers and devices are capable to communicate independently without requiring any computing time from the cpu. Since UDMA mode were introduced the EIDE bus is also capable of some limited form of bus mastering with much more demand for cpu resources. This results in a noticeable speed disadvantage which is made even worse because EIDE devices usually feature lower rotational speeds, higher seek times and less cache storage than their SCSI counterparts.

Another outstanding feature of the SCSI architecture is the possibility to attach and run internal and external devices on one single controller, and in fact between seven and fifteen devices depending on the SCSI standard used. This permits one to construct rather complex device combinations, eg. consisting of two internal hard disks, an internal CD reader, an external CD writer and an external scanner, without much trouble. The bus only requires that each device uses a unique ID number (these ID numbers between 0 and 15 are freely assigned to SCSI devices by means of jumpers, dip switches, pushbuttons or turn-switches, without any requirement for a particular order).

In addition we can even 'sub-address' up to 7 more devices per ID number as Logical Unit Numbers (LUN's), so that in large storage systems like the IBM 3516 several hard disks can be combined to act as a single drive. The usage of LUNs, however, has to be supported by the operating system.

The EIDE interface on the other hand features two ports capable of attaching up to four devices only by using this weird so-called master/slave operation mode, and all of them have to be internal devices. This means that scanners and tape drives meeting professional demands are simply out of the question. This master/slave operation mode only allows for some limited bus mastering, and the overall performance of the EIDE system is always determined by the slowest device. So it doesn't make much sense to, in a worst case example, combine a UDMA-100 hard disk as master device on the first EIDE channel with a double speed CD reader as a slave.

Another advantage of the SCSI technology results from the possibility to run more than one controller in a single PC system.For example, by using standard SCSI controllers from Adaptec we can run up to two controllers at the same time (see ²) which, without considering the possibility of using the not so widespread LUN feature, gives us the opportunity to attach up to 14 (Narrow SCSI) or even up to 30 devices (Wide SCSI) to a single PC. IBM's PS/2 systems are even capable of dealing with up to four controllers, so nothing prevents us from even attaching large external "juke boxes" containing a vast number of hard disks.

Another big SCSI advantage compared to the cheap EIDE standard is its flexibility: While only hard disks, CD/DVD readers/writers and some cheap streamers are offered as EIDE devices, we also have SCSI scanners (mostly for professional use), juke boxes (as mentioned before), DDS-, Exabyte-, DAT-, MO- and several other mass storage devices all of which meeting high quality standards. That's where you can simply put EIDE aside.

Last but not least I have to point out one more big but mostly forgotten difference between the two mass storage standards. The most important device type attached to these busses are hard disks, and it shouldn't remain unnoticed that all those cheap EIDE disks available today are nothing more than mediocre devices at best (in spite of things like UDMA-100 and other similar marketing messages) if their mechanical quality and performance capabilities are concerned. While the fastest UDMA hard disks available spin at 7200 RPM and (with only few exceptions) contain from 512 Kb up to 2.048 Kb cache storage only, we are used to current SCSI hard disks spinning at 10.000 RPM since long time ago, und usually featuring some 4.096 KB, 8.096 KB or even 16.384 KB of cache storage. Furthermore a high-grade SCSI hard disk of today is equipped with a liquid damped head mechanism, and it's storage disks are usually made out of glass substrate. Manufacturer specifications for SCSI hard disks state a very high reliability (expressed in much larger MTBF intervals than those of EIDE devices), meaning that the guaranteed number of start and stop operations of the spindle between any two hard disk failures is more than 100.000 while EIDE devices only manage an average of 30.000. That's why in professional environments dealing with mission-critical data the motto should always be: Don't use EIDE! With the adoption of the forecoming Ultra-4-Wide-SCSI standard and the announcement of the first hard disk prototypes spinning at 15.000 RPM which comply to this standard by IBM and Seagate there shouldn't be any more doubt about the enormous potential of the  high-grade SCSI bus system!

The variety of SCSI standards

It has to be amended that all Low Voltage Differential SCSI systems (LVD) allow for cable lengths up to 25m if there are only two devices (the controller and one single disk) attached to the bus.

The differential and LVD standards are designed for symmetric SCSI (in contradiction to the 'older' SCSI-1-, SCSI-2- and SCSI-3 interfaces which run asymmetrically) and thus are incompatible to the other standards. To still be able to use them with the other standards a voltage level converting device is required, but such a converter is normally integrated into a modern hard disk's electronics already. So if you're using an IBM SCSI hard disk you can easily set the operation mode to either "SE" (= Single Ended) for asymmetrical SCSI or "LVD/Diff" (= Low Voltage Differential/Differential) for symmetrical SCSI by applying a jumper.

II. Plugs and controllers

The variety of plugging standards

Flatbed scanners usually use 25 pin D-Sub plugs while external streamers, CD readers and CD writers are more commonly equipped with centronics plugs or pen plugs. Internal SCSI-2 devices are connected via 50 pin cables whereas devices belonging to the wide SCSI standard feature 68 pin connectors. SCA plugs are exclusively to be found in high-grade hard disks.

Types of controllers

Due to the many different possibilities of SCSI controller usage it's quite impossible to give a universal recommendation towards what type of controller you should buy, but there still are some general advice that you should follow: if the SCSI controller will be your only interface to run mass storage media, I recommend buying some up-to-date model. In doing so, you should watch out for the following features:

• The controller should contain a SCSI bridge and offer two connectors (50 and 68 pins) to make it possible to attach either SCSI-2 and SCSI-3 devices without any additional hardware.
  
• Under any circumstances there has to be a bios on the controller making it capable of booting from any given hard disk.
  
• It should have an external 68 pin connector for the attachment of external devices.
  
• It should feature automatic termination (like most modern controllers do).
  
• If your PC power supply is out of power cables you should make sure that there's one for internal devices integrated into the controller.
  

1. The controller's bios (if there's a bios on it at all) should be stored in a socketed chip so that, since a SCSI bios is only needed on a controller that has a boot device attached to it, it's possible to be removed.
   
2. A secondary SCSI controller is usually used to attach some slower devices like scanners. So there has to be an external connector on it; in most cases SCSI-2 is enough for this purpose because even modern scanners are unable to use the full potential of a SCSI-2 controller.
   

Most of today's SCSI controllers sport an automatic termination feature, which is triggered when the controller realizes that it's the last device on one side of the bus. When dealing with older controllers you should make sure that they contain some special sockets for terminators. On older controllers these are normally to be found as some rectangular structures painted in light yellow, and they're usually located near the external connector.

Additionally you should use high-grade cables only, especially with the newer Ultra-2- and Ultra-3 interface cards. Cheap connection cables that do not adhere to the given standards might cause extreme bus disturbances. The external cables should also be of the high-quality type (like using gold plated connectors and appropriate electric shielding), because bad cables can reduce signal quality significantly leading to a decrease of the overall length of the bus. This gets especially important with a mix of internal and external devices, because the overall bus length is determined by the length of all internal and external cables used. And while we're at it, I strongly recommend the use of active terminators. Though SCSI-1 and SCSI-2 systems can optionally be equipped with passive terminators too, any increase in device performance and cable length can lead to problems with these standards already. With SCSI-3 and above active termination is required anyway.

The 'Crème de la Crème' - RAID systems

1. Level 0: This level, also referred to as 'striping', doesn't really offer any advantages in data integrity compared to a standard hard disk subsystem. When RAID-0 is used the controller splits all data into packages and stores them on several physical devices acting as one large device. The main advantage of this level, which is augmented by the bus mastering feature of the SCSI bus, is its extremely high speed advantage when carrying out read operations, because data can be read from all devices of the RAID subsystem at the same time which theoretically results in a data transfer rate that is the sum of the transfer rates of all participating devices.
   

   But this gain of performance of course depends on some other circumstances as well! Mediocre file systems like, for instance, FAT32 or NTFS - both of which are standard filesystems in some Microsoft products - reduce the possible performance advantage because they automatically lead to file system fragmentation. OS/2 Warp's HPFS makes much more efficient use of such a hardware, as does HPFS386.
   

   The greatest disadvantage of RAID-0 is it's imminent data insecurity: if one of the participating devices fails or if we get read or write errors all data is lost and can only be recovered with extreme effort and on all RAID devices, if it can be recovered at all, that is.
   

2. Level 1: This RAID level, also called 'mirroring', implements some simple data redundancy by, as the name already implies, mirroring all data on several devices, so that every data item exists twice.
   

   The greatest advantage of mirroring is it's extremely high data integrity: if one of the hard disks fails all data operation can be seamlessly carried on by using the remaining disk. Its disadvantage is its large amount of slack space - while on a RAID-0 subsystem the storage capacity is a sum of the space of all participating hard disks, with RAID-1 the maximum storage available is determined by only one of the disks because the mirror disk is only use for redundancy purposes. What's more, in a RAID-0 system there will be no performance gain at all.
   

3. Level 2: This level is quite unimportant and hardly ever used in practice. It mainly differs from level 0 by data dislocation on the bit level. Integrated parity check and error correction algorithms make for a higher reliability of data storage, but transfer rates are being reduced by the complexity of parity checking.
   
4. Level 3: This level uses a reduced data block size of 1 byte only and also features parity checking. Furthermore all parity data resides on a special dedicated hard disk. As a result of this administration effort level 3 features an extremely bad performance when data is written, which is why this RAID level wasn't widely used either.
   
5. Level 4: This level in principle works like level 3 with the exception that data is being stored in blocks rather than byte wise. This level also requires a dedicated hard disk for parity data. Level 4 is hardly ever used in real-life environments.
   
6. Level 5: This very often is used RAID level stores data quite like level 0, but parity information are stored with the data thus eliminating the need for a dedicated parity hard disk. RAID level 5 features a very high performance and still good data integrity but comes rather expensive by requiring special controllers. Most controllers available support RAID levels 0 and 1 only. Another RAID 5 advantage is it's great storage capacity because, like with level 0, the overall capacity of the array is the sum of capacities of all participating hard disks.
   
7. Level 10: The so-called level 10 is merely a combination of levels 0 and 1. This results in a good performance like level 0 combined with the high data integrity standard from level 1. But since level 10 consists of two stripe sets and doesn't add any technical enhancements to levels 0 and 1 there still are some disadvantages: a level 10 system has to consist of at least 4 harddisks because a stripe set is always made of at least 2 hard disks. Because these stripe sets are also being mirrored we lose a lot of storage capacity because with four hard disks in use we only get the capacity of two. This rather uneconomic, cheap solution is mainly to be found on the newly introduced EIDE-RAID controllers which only offer RAID levels 0, 1 and 10, whereas controllers out of the more professional SCSI areas normally implement RAID-5, the most secure and most functional solution in every aspect.
   

Mass storage devices

The attachment of hard disks, however, might lead to some special requirements: Since hard disk technology always went along with the evolution of SCSI itself or has even raised new standards, we have to deal with the same variety of hard disk types on the market as with controllers. Scanners, however, are mostly distributed with SCSI-2 interfaces, and because of their limited speed a SCSI-2 controller is absolutely fast enough for them. Same goes for CD readers/writers - most of them are manufactured as SCSI-2 devices.

It's no problem to have an Ultra-3-Wide controller with an Ultra-3-Wide harddisk and a second one with a SCSI-2 interface only attached as long as the controller features a bridge and appropriate 50- and 68-pin ports. The controller simply runs each disk with its maximum velocity, so the slower disk will not end up being a bottleneck for the overall performance of the SCSI bus. Controllers featuring only 68-pin ports are an exception though, because the slower SCSI-2 disk now will have to be attached to the much faster Ultra-3-Wide bus. Keep in mind that, in such a case, you'll have to use a so-called interface converter to make the different plug layouts match each other. Such a converter is a small circuit board which provides compatibility between these different plug layouts. On the more sophisticated converters we will also find some jumpers, because they have to be tuned to the SCSI ID of the hard disk they're attached to.

If you plan to use a Ultra-3-Wide-SCSI hard disk with a Fast-SCSI-2 controller and a second hard disk which is SCSI-2 - no problem. Again it's time for an interface converter, which should be available in electronics stores for about 15 EUR. Please note though that the Ultra-3-Wide hard disk will no longer offer its full performance here, because the controller defines the maximum data transfer rate. This means that the Ultra-3-Wide hard disk will only run like a Fast-SCSI-2 model. Still you will experience some performance advantages compared to an ordinary SCSI-2 hard disk, because modern Ultra-3-Wide disks feature lower seek times, bigger caches and higher spin rates.

One very unpleasant source for configuration errors which can give SCSI enthusiasts a hard time is the possibility to attach a SCSI-3 hard disk to the internal 50-pin socket of a SCSI-3 controllers. This one is somewhat  perfidious: Since the hard disk negotiates the communication mode with the controller and, with both devices being Wide-SCSI, will switch to 16 bit mode, the hard disk will be recognized and its SCSI ID will be recorded correctly but afterwards it won't be addressable any longer. This behaviour results from the limitations of the 50 pin cable used, which is only capable of transferring the lower 8 bits of data so that every second byte will be lost. To provide a circumvention to this problem most SCSI-3 hard disks offer a jumper to switch a so-called "Target Initiated Wide Negotiation" or "Target Initiated Synchronization Negotiation" on or off. In our special case we would have to put this jumper on or off while in the controllers BIOS we would have to prevent Wide Negotiation  for the hard disk's SCSI id. People owning SCSI hard disks from IBM have to watch out for some more potential trouble, because IBM sells two different flavors of their disks - simple types (eg. DCAS, DDRS, DNES) which are designated for "Low end workstations, low end file servers" and "desktop personal computers" and some high-grade types (z.B. DRVS, DGVS, DDYS, DGHS) with the brand name "Ultrastar". Ultrastar models primarily differ from the other ones by having bigger internal caches, higher rotational speeds, more sophisticated mechanics and shorter seek times. In addition the high-grade models get their "Target Initiated Synchronisation Negotiation" jumper set to off by default, whereas the entry level models usually have it set to on. So, with an IBM hard disk you'd better read the manual!

Another source for errors could be a mixture of SCSI-2 and differential Ultra- x-Wide-SCSI. In such a case you'll have to definitely and absolutely make sure that some special setting gets activated on those Ultra- x-Wide harddisks: Since they are addressed as LVD devices, you'll have to set a jumper on these hard disks on a setting called "Force SE" when they are used in asymmetrical environments (SCSI-1, SCSI-2, SCSI-3 etc.)! If you omit this, the hard disk might not recognize that it is being run in an asymmetric mode and may be destroyed by the wrong signal levels resulting in this case! So make sure that this "Force SE" (SE means 'single ended') jumper has been set before the disk is mounted! This setting also has to be applied to an interface converter by the way, else you'll experience incompatibilities that could lead to system failure or even hard disk destruction!

Since many Ultra-x-Wide hard disks sport SCA interfaces, you'll have to take care that a given interface converter for SCA offers the appropriate power supply for the disk. If you omit the hard disks power supply, it disk will not run. This could lead to system crashes.

Another exception would be the usage of some ancient SCSI-1 harddisk with a brand new Ultra-3-Wide harddisk on a controller of the latter type. Technically this combination is valid, but most SCSI-1 disks don't implement the controllers disconnect-/reconnect. command set, either not at all or in some limited form only. Therefore, just like EIDE, such a disk might lock the bus during transfer operations. Because of the  low speed of such devices compared to the modern ones they may then block all other devices for a considerable amount of time. If, in spite of such effects, such an old device has to be used, you should attach it to a controller of it's own, just like you would attach a scanner. That controller may by all means be of some older type as well. This guarantees that modern mass storage devices with their vast performance don't get affected by some old components and that overall performance drops to the level of a machine which is some ten years old. Even with a SCSI-2 controller the negative impact of such an old device is perceivable thus requiring a controller of its own.

So, after this quick introduction to some theoretical, historical and user specific basics we will deal with some more practical stuff and show some SCSI configurations in the next issue, where we will also discuss OS/2 Warp specifics in detail.

References: 1 The ESDI interface, like its ancestors, was limited to internal devices as well and could only address up to two hard disks. 2 See: Dorle Hecker/Hans-Jürgen Götz: OS/2 WARP Version 3 Integrationsplattform, Franzis-Verlag GmbH, Poing, 1995, S. 53. 3 Detailed comparisons between EIDE and SCSI can be found at (German magazine): Stiftung Warentest: PC aufrüsten. Für Einsteiger und Fortgeschrittene (PC upgrades for beginners and professionals), Berlin 1999, S. 150 ff. and (German book) Elke Kleinknecht/Bernd Rohrbach: OS/2 Praxislösungen (Practical solutions for OS/2) , Band 1, Teil 2/2.6, Interest-Verlag GmbH, Augsburg, September 2000.

[Feature Index]
editor@os2voice.org
[Previous Page] [Newsletter Index] [Next Page]
VOICE Home Page: http://www.os2voice.org