8.      Enhanced ATA Interface - some unsorted thoughts only ....

        For some reasons, there are real wars going on between the
        hard-nosed SCSI and the identical ATA guys. Mostly i try to
        stay out of this discussion, but to get a better view about
        SCSI, a few thoughts should be added about ATA.
        This is based on IDE's and ATA's technical data, so the facts
        are correct. However, my personal preference is SCSI, so keep
        in mind that opinions here are my personal ones.

        ATA (AT Attachment), also often called IDE (Integrated Device
        Electronics), has a few principal limitations - like every
        other interface, also including SCSI. ATA's basic limitations
        as i would see them, including EATA, are the following:

        - only usable with ISA bus or an ISA-to-something-else bridge
          So it seems to be somewhat limited to the PC architecture,
          although there are a few exceptions like Apple's basic
          "Performa" systems.
        - only two disks can be attached per ATA channel, but two ATA
          host adapters are possible in one system - some new VL/ISA
          adapters have this feature, addressing up to four devices
          - if the BIOS doesn't refuse to play this game
        - limited device types  - disk drives
                                - CDROMs
                                - tape drives
        - Host PC locked during the transfer time
        - address range is limited, ATA at 504 MByte, EATA at 8.4 GByte
        - Data rate is limited, - ATA at 8.33 MByte/s,
                                - EATA now at 13.3,
                                - EATA soon at 16.6 MByte/s
        - external devices are virtually not possible.

        ATA's concept isn't very professional in using the range of
        head, cylinder, and sector adress bits. Although DOS could
        address up to 8.4 GByte disk size and ATA could address up
        to 130 GByte, the combination of both schemes and the PC's
        BIOS limitations - neccessarily taking the smallest number
        for each parameter - ended up in 528.482.304 addressable
        bytes - 504 MBytes.
        With the - slightly chaotic - actual situation with vendor-
        specific address extension schemes in mind, ANSI adopted an
        addressing scheme from the Small Form Factor committee's
        draft standard SFF-8019. It uses 16 heads, 16383 cylinders
        and 63 sectors to achieve the 8.4 GBytes address range DOS
        can handle.
        Although the 8.4 GB limit seems unproblematic, especially
        with DOS still the standard operating system, i'm not too
        happy with it for two main reasons:
        First, there are drives in the market with 9+ GB - i saw
        more than one question like "how can i format the Seagate
        9GB disk drive for DOS?" at CompuServe in the last months
        - too much to push this thought aside. Besides, bigger
        drives are announced.
        Second - remember the IBM PC with its 640 kB limit? Everyone
        thought "640 kB - wow, i will _never_ need this much" - two
        years later the first memory expansion boards came out ....
        A size limit that's already visible with actual devices is
        too tight for me to be happy with it - it raises the question
        about a technology's lifespan.

        Now, if you compare EATA's maximum specs, it looks like the
        difference to SCSI is smaller than ever, with some spare
        capacity. Unfortunately, the major drawbacks still apply:

        - programmed I/O on ISA systems is mostly limited to about
          4 MBytes/sec, everything over this needs EISA, PCI, or VL
          bus systems. DMA busmaster mode isn't possible.
        - During the transfer, the host PC is tied up while waiting
          for the transfer completion, regardless if the transfer
          bandwidth is used or not.
        - With most actual host adapters, the slowest device sets the
          pace. Although individual timing for every device at an ATA
          channel is possible, most actual adapters don't support this.
          So, with a disk drive capable of PIO mode 3 (11.1 MB/s) and
          a second, older drive with only PIO mode 0 (3.33 MByte/sec),
          you can set the adapter only to PIO mode 0 - crippling the
          faster disk drive in the name of the holy compatibility <g>.
          However, newer EIDE adapters should support it.
        - The official cable length limit is 18" - with a hint in the
          spec that this is the absolute maximum including the on-board
          bus length, and that less is desirable due to transfer
          problems even inside the legal range. This virtually outlaws
          external devices except in _very_ careful designed - and
          expensive - controller and cabling setups.

        The first problem of course applies also to SCSI host adapters
        working with programmed I/O - but there are _lots_ of SCSI
        busmaster adapters in the market, and the PIO adapters are
        mostly present in the lower market segments.
        The second problem is addressed by the Disconnect/Reconnect
        mechanism in SCSI - while the disk seeks, the CPU can do
        something useful - read a sector from the CDROM, calculate
        fractals or so on ... <g>  I couldn't find anything similar
        in EATA, but i would like to hear that i'm wrong....
        Third, SCSI devices are independent in their performance -
        assumed that the host adapter can set his maximum data rate
        per device (usual with actual adapters).
        Actually, this leads to a better real-life throughput with
        SCSI, compared at similar burst transfer rates. For the same
        reason, CPU utilization with CDROMs is usually lower with SCSI
        CDROMs, as it is possible to unload some work to the SCSI host
        adapter - with ATA, the CPU has to do _everything_ on its own.
        The cable length problem is not as crucial with SCSI. A
        properly terminated SCSI bus will work reliable inside the
        specified length limits.

        There are some intelligent ATA adapters, of course - mostly
        cache controllers with internal intelligence. They make things
        better, but can't bypass the PIO problem without leaving the
        compatibility path; additionally, they eat up most of ATA's
        cost advantage.

        Those thoughts lead me to think that SCSI will stay here, with
        even higher market shares, as device development demands faster
        interfaces and new devices shine up.... ATA will still have the
        higher market share, as lots of people buy at the lowest bidder
        - not likely a SCSI vendor. Also, the 'standard' SoHo user might
        not notice a difference in his daily work - heck, with two
        fairly comparable systems, i don't notice a big difference,
        unless it comes to attaching devices - the 'standard' user with
        one disk drive, one CDROM, and a floppy or ATAPI tape clearly
        will question the benefit of SCSI for him.

        Also, to get real benefit from EATA, the PC needs an extended
        BIOS. The SFF - mainly through its member Phoenix - defined such
        an extended BIOS. This includes basic Plug'n'Play functionality,
        some major functions are slightly changed - the disk parameter
        table gets a 16 Bytes extension, Int 13h functions are extended
        - to allow booting from every attached hard disk, for example,
        and so on ....

        To summarize my opinion - EIDE is ok for the usual system.
        If you are happy with IDE, you will be happier with EIDE - if
        ( a _big_ IF ! ) your system is _fully_ EIDE compliant.
        This means, you need the complete EIDE set, consisting of:

        - an EIDE controller.
        - an EIDE compliant BIOS or BIOS extension.
        - full EIDE disk drives

        The drawbacks are:

        - if you mix with older drives or if you have a single non-EIDE
          component, EIDE is BS - you won't benefit from EIDE over IDE.
        - ok, EIDE is better than IDE. But what's the point in selecting
          a new technology if a better system is on the market already?
          Of course, this is a personal opinion <g>...

        More precise information about ATA/EATA can be found also on
        the SCSI BBS - see some connections <g> ?  But seriously,
        ATA/EATA also is an X3T9.2 project, so the EATA development
        will continue over time.

