6.      Typical SCSI devices

6.1.    Hard disk drives

        SCSI was always a high-end interface for hard disks. For it was
        expensive, it naturally showed up only in high-end disks. At the
        moment, you'll see that all "state of the art" hard disks use
        SCSI and the generation before (in capacity/technology terms)
        gets equipped with IDE/ATAPI interfaces for the mass market.
        The point where this technology transfer comes in changes
        slightly - now there are 1GB+ ATAPI and EIDE hard disks, but the
        2GB+ range is still - i think - a SCSI domain.

6.1.1.  RAID drive arrays

        RAID is a technology that's nearly "married" with SCSI. RAID
        is a method to combine two or more disk drives to a bigger
        logical drive with or without redundancy/fault tolerance.
        The acronym RAID (Redundant Array of Inexpensive Disks) tells
        the main reason why it came up - either combining cheap disks
        to sell a disk system for much more money than the disk drives'
        cost <g>, or combining drives to get a bigger drive that was
        possible with a single drive.
        Now, RAID is primarily a security option, for it's actual used
        types add redundancy to the disk system to be able to swap a
        defective disk without using data.

        RAID exists in several levels ordered by numbers;

        RAID 0, called "striping", combines two or more disks to a big
        logical disk drive. The data is distributed between the disks
        by a "striping factor", that means, data are separated by blocks
        of, let's say 32 kByte, between the disks. This could, not
        neccessarily must, give also better performance with big files,
        because the data come from parallel drives, thus the mechanical
        access times are reduced.
        Raid 0 gives no security plus.

        Sometimes you'll see "disk spanning" advertised as a RAID 0
        feature. Spanning isn't RAID, though, as it is just combining
        two or more disks to a bigger logical one by adding up the
        sector numbers without striping. Spanning was mostly used in the
        past, when single disks didn't get over 1GB.
        Spanning was easy implementable in driver software. SpeedStor's
        abilities, for example, also include disk spanning.

        RAID 1, also called "disk mirroring", adds no capacity, but
        enhances security by reading/writing the same data from/to two
        disks. With an intelligent two-channel controller, this can
        sometimes lead to a small performance plus on read operations,
        for - if all disks can read and write simultaneously - the first
        disk completing the read process sets the "time mark". Write
        operations normally become slower, as the write must be done on
        all drives before it's completed.
        In real world, it's normally a speed brake, especially its
        software variants.

        RAID 2 uses a dedicated drive for error-correcting information.
        The high redundancy of RAID 1 (2 GB disks for 1 GB data) was
        reduced by using Hamming-codes for the correction data to about
        40%. RAID 2 needs ECC error correction built-in in all disks.
        In the RAID 2 model, striping is implemented at bit-level, that
        means, bit 0 goes to disk 0, bit 1 to disk 1 and so on...
        However, i never saw or got info about a RAID 2 system, for they
        would be very expensive to implement, if ever. In fact, it was 
        never implemented.

        RAID 3 and 4 consists of at least 2 data drives and a dedicated
        ECC data drive. Data is striped between the disks, typically in
        byte-packets and XOR combined for the ECC drive. Typically, the
        performance is good with large file reads, worse with small file
        accesses or many writes on a network server.
        RAID 4 used very high striping factors, thus giving better
        performance than level 3, but still somewhat slow, for the ECC
        drive still has to be used on _every_ disk write access.

        RAID 5 distributes the ECC data between all drives, so the
        bottleneck of level 3 and 4, the dedicated ECC drive, vanishes.
        So, disk accesses can overlap, thus giving better performance.
        Also, the capacity loss through redundancy drops to 25%.

        RAID level 5 seems to be the last "official" RAID level number,
        although a level 6 was defined years later by the RAID Advisory
        Board (RAB). According to what i could find out, RAID 6 is
        basically a Level 5 with the added capability of asynchronous
        and cached data transmission. If you have more precise
        information about RAID level 6, please let me know....

        If you read level numbers like "RAID 6", "RAID 7" or "RAID 10",
        you should insist of a detailed explanation about what they mean
        with the level numbers. Typically, it is a vendor's proprietary
        RAID implementation, mostly added from a combination of two
        "official" RAID levels.

6.1.2.  AV disk drives

        In the last months, the "AV" drives came into the market that
        are designed specifically for the Audio/Video market.
        For AV needs a continuous data stream more than the last
        percents in performance, those drives mostly have a different
        firmware optimized for AV operation. The most common changes
        include the avoidance of thermal recalibration while read or
        write operations take place or are pending. One other point is
        the inclusion of spare sectors in each track, so that a replaced
        sector doesn't cause head movement. For this - and some other -
        reason the AV drives are mostly a bit slower than the similar
        "standard" drive, and have a lower capacity. In general, these
        disks should be used only for machines where this "isochronous"
        behaviour is absolutely neccessary. In all other cases, they
        are a waste of money compared with fast standard drives.


6.2.    Removeable disk and Magneto-Optical drives

        All removeable devices i used myself started as and still are
        SCSI devices. The only other devices i know are Syquest's Puma
        IDE drive and an older Sony MO with an ESDI interface and a
        special interface card. Also, there should be an ATAPI version
        of IOMega's Bernoulli-Box, and the newer drives tend to be
        available with ATAPI and SCSI interfaces. Even Fujitsu
        announced an ATAPI-MO drive recently.

6.2.1.  Bernoulli and Syquest drives

        Bernoulli-disks are named after Daniel Bernoulli, a Swiss 18th
        century mathematician. The Bernoulli-effect is basically that:
        between a fixed metal plate and a flexible rotating medium, it
        develops a very thin and highly compressed air-cushion that
        keeps distances low but surely avoids contact. So, the head-to-
        media distance of a Bernoulli disk is lower than in a hard disk,
        without the shock sensitivity of a hard disk. If this air flow
        is distorted by dust, dirt, power loss, mechanical shock or
        whatever, the medium simply "falls off" from the head plate
        without damage. Bernoulli drives are available up to 150 MB,  
        and the 150MB Multidisk also reads and writes the older disks.

        The new 100 MB ZIP drive is IOMega's new low-cost approach. It
        is based on the ATOMM technology, a media developed by Fujitsu,
        but supposedly not fitting in Fujitsu's MO-centered philosophy.
        The ZIP seems to be a nice drive - cheap, fast, cheap media.
        Let's see when - and if - IOMega comes out with an internal
        3 1/2 inch version....

        Another new announcement from IOMega is the "Jaz" drive - a
        removeable, ultrafast (up to 6.7 MB/sec sustained data transfer
        rate, according to the announcement), and 540 MB or 1 GB disks.
        If it really comes at under $500 for the drive and at $99 for a
        1 GB medium, i would expect it to become a winner. Let's see...

        Syquest drives are working basically like "normal" hard disks.
        The medium is a - removeable - metal plate, heads and drive
        spindle remain in the drive. Working with the same technical
        principles, they also share the hard disks problems - Syquest
        drives are shock sensitive.
        With the old external 44MB and 88MB drives, i remember that the
        drives were very loud, but possibly mostly due to its fans.
        The 3 1/2 inch drives with 105 and 270 MB seem to change this;
        the 270MB drive is one of the most appealing devices i saw -
        fast, quiet (compared to the older ones) and small.

        There are some ongoing technical and legal struggles between 
        Syquest and Nomai, a former Syquest distributor and by now
        a competitor - Nomai sells own-built Syquest clones and has
        advertised a 540 MB drive for the near future.

        Also, there are rumours about a new 135 MB Syquest, but i don't
        have more about that.


6.2.2.  MO drives

        MO's all work basically the same way:
        For writing, the magnetizable layer in the medium is heated with
        a laser beam up to its Curie-temperature (approx. 150 Celsius).
        Then the magnetic write head magnetizes the track according to
        the data. Every polarity change is a bit.
        Reading is pure optical. A low-energy laser beam gets reflected
        by the data layer and - according to the magnetizing - gets
        polarized by the medium. This is called the "Kerr-effect" and
        is used to get the data bits from the reflected beam.
        MO has two key points - slowness and reliability. Reliability -
        the media usually are certified for ten to 30 years, also the
        media are not sensitive to magnetic fields, water, radiation
        and mechanical shocks (at least as long as the media itself
        isn't damaged).
        The - relative - slowness is due to the principle; every write
        process needs two turns of the media - in the first turn, old
        data are erased, in the second turn the actual data are written.

        The future seems to belong to the 3 1/2 inch devices - an ISO 
        standard with about 640 MB is scheduled for the end of 1995,
        and higher capacities up to 2 GB are in discussion. Anyway,
        the MO drives are the only ones of the small removeable
        devices that aren't proprietary.

6.2.3.  MD-Data

        Sony tries to push their audio MD in the data market. The
        MD-Data is a 2 1/2 inch medium with 140 MB capacity. At the
        moment, the drives aren't widespread, if even available, but
        there are a few interesting approaches that come with the MD.
        The major advantage of MD-Data against standard MO is its
        "Direct Overwriting" capacity. At the moment, the drives are
        very slow (150 kB/sec), but this should change soon as the
        technology matures. Also, the capacity for the data-mode
        drive may increase soon.
        Sony and Microsoft - and apparently a few others - are
        developing an universal file system for the MD-Data, that
        should enable transparent exchangeability between different
        platforms and operating systems.

        Whatever will happen to MD-Data - after the introduction hype
        in 1994, the last months were _very_ quiet about it....


6.2.5.  WORM drives

        WORM (Write Once/Read Many) drives are very seldom used - for
        each surface spot is writable only once, the media aren't
        reusable and therefore are used only for archiving purposes
        where security against manipulation is a factor.
        Actually, there are two main systems - phase change and simple
        "toasting", as i would call it. In the phase change system, an
        empty medium has an amorphous active layer, and writing the
        medium with a high energy laser beam "changes" the written
        portions of this active layer to a crystal state ("phase") with
        a different reflective behaviour.
        This method is basically irreversible without destroying the
        medium completely, but lately became advanced to an MO variant
        (see "Phase-Change MO").
        The "toasting" is simpler - the medium is "burned" by a laser
        pulse and permanently deformated, thus achieving a different
        reflection behaviour.

        WORM media are expected to have slightly shorter life cycles
        than MO media, but 30 years expectancy doesn't seem too bad...

6.2.4.  Phase-Change MO drives

        Lately, Panasonic and Epson introduced the Phase-Change optical
        drives. A more modern variant of the phase-change process used
        in WORM drives is reversible and thus seems nearly ideal for
        the purpose...

        The main difference to standard WORM drives is that the active
        media material is a highly reflective crystalline material
        that allows deleting/rewriting - voil, a WMRM <g>. Compared to
        MO, the rewriting process happens in one revolution, without a
        separate delete process, thus making writing faster.
        Both drives i saw advertised (Panasonic and NEC) were combined
        with a CDROM reading optic, so that they can be used with either
        a MO cartridge or a CDROM - a nice feature, if you need both
        media and don't have enough room for both, or if you don't have
        any drive yet.


6.3.    CDROM drives

        A word first: much more information about CDROMs than here is
        collected in Kevin Kelly's LASER3.TXT file. He wrote the most
        complete CDROM introduction and hint collection i ever saw -
        give it a look. It is available from CompuServe's CDROM forum.
        Of course, it may be named LASER4.TXT or so, when you look <g>.

        SCSI CDROMs mostly share the CDROM Common Command Set. This
        means, all these CDROM drives work with a standard driver, at
        least in their data mode. However, the audio commands are not
        standardized, so you might encounter driver problems here,
        when installing a brand-new drive with older drivers. All
        drivers i know default to support data-mode only in this case.
        NEC CDROM drives are my special enemies <g> - if they work,
        mostly they are good drives, but there are a few drawbacks...
        Older NEC drives can be a bit difficult, especially the old
        portable CDR-35 and CDR-36 models. Their SCSI implementation
        is described best as "daring" - simply "bad" isn't enough here.
        I got my old CDR-35 working _only_ with a Trantor MiniSCSI
        parallel adapter and only, if it was the _only_ device on the
        bus, but not with various Adaptecs and also not with the FD
        adapters i could test with.
        The CDR-73/74/83/84 drives mostly work, if "SCSI parity" and
        "Synchronous Negotiation" are disabled for their IDs or the
        whole bus.
        Also, some NEC drives don't provide termination. Normally
        this isn't a problem if the CDROM is the only device on the
        SCSI bus, for it _mostly_ works this way. If you need a longer
        SCSI cable, the problems start.
        Keep this in mind if you integrate a NEC CDROM in your existing
        SCSI setup - some models can't be on the end of the chain,
        except if you have an internal pass-through terminator for them.

        One customer has a CDR-500 (3x) that finally works flawlessly
        with a 1542CF and a Seagate 1GB disk - a very nice drive;
        however, there were two firmware upgrades neccessary before,
        and i heard about trouble with some firmware releases from
        many different sources.
        The latest NEC i got was a CDR-210 - a cheap double-speed drive
        manufactured in 1994. You guess it - no parity, no termination..
        There are two CDR210 - a model without parity and termination
        ability, and an OEM version CDR-210P, which supports parity.
        Again, you guess what version this customer got <g>.

        Sadly, the story seems to continue with the new 4x drives -
        they doesn't seem to be better. One of my office customers
        bought a 4x with firmware 2.2. Because the system for this
        drive wasn't delivered, i tried it on my systems. If it works,
        it's a very nice drive - fast, very good error correction.
        The trouble i have with the 4x is, that it mostly locks up on
        a warm reboot, if a disc is in the drive - the standard case
        on my system. But switching it off, on again and _fast_ (it
        won't react after the first two seconds) pulling the "eject"
        lid allows me to reboot my system without lockup....


6.4.    Jukeboxes and other media changer devices

        Jukeboxes for all types of media (CDROMs, WORMs, MOs, DAT
        cartridges) are generally available as SCSI devices only, as
        SCSI is the only standardized interface that supports them
        in PC architectures. Either each medium gets a dedicated ID
        that can be addressed by the PC or the different media are
        combined to a single structure.
        Of course, the various Mainframe interfaces are also supported
        by devices like WORM changers - these devices originally came
        from this "Jurassic park" world into the lower-end system
        worlds like workstations and PCs. But in most cases i saw
        there are only bridge-controllers in the system to connect a
        SCSI device to the proprietary mainframe channel.


6.5.    Tape drives

        Professional tape drives started - after some proprietary
        interfaces - with the QIC-36 and QIC-02 interfaces, then fast
        tended towards SCSI. For the lower-end market, the QIC
        consortium defined QIC-107 and -117, the floppy-disk based
        interfaces for the QIC-40 and QIC-80 mini-cartridge tape
        drives.
        QIC-117 strongly depends on the PC's floppy disk controller's
        timing and signaling and therefore isn't very portable to
        other architectures.
        High-end tapes like bigger QIC with 500MB+, 4mm and 8mm DAT,
        DLT and so on generally aren't available other than SCSI and
        maybe a few specific interfaces from the mini and mainframe
        world like Pertec. IBM's 3480/3490/3590 drives usually are
        available with SCSI and ESCON interfaces. SSA should follow
        soon, at least for the IBM drives.
        The capacity limit for non-SCSI devices may slightly change in
        the future, for the new QIC-30xx tape drives and - based on
        them - newer tape concepts like 3M's "Travan" promise new tape
        drives with up to 4 GBytes capacity - we'll see what happens
        here. However, there seems to emerge a whole bunch of new,
        incompatible (of course <g>) tape formats, and some of them may
        come up with SCSI devices, too (they still appeared - Tandberg's
        Panther Mini drive - 1 GB with QIC-3040 Wide - is an example).

        Actually, the tape drive world changes rapidly in the world of
        small form factors - the new high capacity QIC-30xx, QIC-Wide
        and TRAVAN formats emerge for all interfaces, newer low-cost
        drives appear for the ATAPI interface and the incompatibilities
        become bigger than ever - i finally move new changes to this
        theme to the QIC-appendix (App. D).


6.6.    Scanners and other imaging hardware

        Scanners exist with a lot of interfaces. In the past, serial,
        parallel and IEC interfaces were common besides SCSI. The higher
        resolution and color depth fast decided this interface struggle
        towards SCSI. Generally, professional scanners use SCSI now, but
        keep this history in mind when looking at some older, but maybe
        very good scanner. The SCSI models mostly work with standard
        host adapters, if you find a driver for them. Canon and Agfa, 
        for example, switched to full ASPI support only a few months 
        ago.

        SCSI scanners are often somewhat "incorrect" in identifying
        themselves to the system. My EPSON GT-6500 for example, tells
        me "EPSON SC" as vendor info, and "ANNER GT-6500" as model,
        besides the SCSI-0 (non-ANSI compliant) id it gives to the
        host adapter.
        An older AGFA scanner from a customer of me gives back only
        trash if asked for its vendor and device strings.
        This comes, for most Scanner manufacturers have their own
        pre-SCSI-II command set that they rely on. There was never a
        standardizing neccessity as with hard disks, where devices not
        compatible with the SCSI CCS would have been impossible to sell.
        So, SCSI can be still sort of a mess with most scanner vendors.

        Actually, you can have most professional imaging hardware like
        digital cameras either with a proprietary PCMCIA card or with
        SCSI interface. This also includes a few SCSI-based frame
        grabbers.

