Hardware opinion: Slow SCSI or fast IDE - Netware

This is a discussion on Hardware opinion: Slow SCSI or fast IDE - Netware ; I need to make a short term 3.12 (yes, that's not a typo) server. The existing box is a 486 66mhz 128mb ram with a 2Gb SCSI drive and 10mbit card serving around 8 users. The box only has 16bit ...

+ Reply to Thread
Results 1 to 7 of 7

Thread: Hardware opinion: Slow SCSI or fast IDE

  1. Hardware opinion: Slow SCSI or fast IDE

    I need to make a short term 3.12 (yes, that's not a typo) server.

    The existing box is a 486 66mhz 128mb ram with a 2Gb SCSI drive and
    10mbit card serving around 8 users. The box only has 16bit slots.

    I could try to replicate the above or more easily grab a spare PIII
    350mhz, 256mb RAM 100mbit PCI card with oodles of disk space on the
    IDE drive.

    Although I've always adhered to using SCSI drives for servers I
    suspect that the faster IDE drive with extra cache memory available
    will run rings around the older 486 machine.

    What do you reckon?

    --
    AnthonyL

  2. Re: Hardware opinion: Slow SCSI or fast IDE

    The message <431aaddf.3077224@news.zen.co.uk>
    from nospam@please.invalid (AnthonyL) contains these words:

    > I need to make a short term 3.12 (yes, that's not a typo) server.


    > The existing box is a 486 66mhz 128mb ram with a 2Gb SCSI drive and
    > 10mbit card serving around 8 users. The box only has 16bit slots.


    > I could try to replicate the above or more easily grab a spare PIII
    > 350mhz, 256mb RAM 100mbit PCI card with oodles of disk space on the
    > IDE drive.


    > Although I've always adhered to using SCSI drives for servers I
    > suspect that the faster IDE drive with extra cache memory available
    > will run rings around the older 486 machine.


    > What do you reckon?


    Unlikely to make a blind bit of difference. My first experience of
    netware was 2.15C on an 8MHz 286 box (NEC PowerMate 2, as it happened),
    with 600MB ESDI disk.

    I upgraded the MoBo to a 386SX33 one and saw very little improvement.
    Having upgraded to a 386 cpu based system, I was then able to ditch the
    2.15C and install 3.11 which I quicly upgraded to 3.12 which I was using
    up until a few weeks ago when I replaced it with a Debian based samba
    server.

    After installing 3.12, I regularly upgraded the hardware through 486,
    SKT7 P166, SKT370 Celeron 466 and (briefly) skt370 p3/500 and finally
    back down to an under volted and underclocked skt7 K6/500 (running at
    250MHz) reducing the power consumption to a mere 29 watts with a single
    250GB Seagate drive.

    In all that time and in spite of upgrading from the on-board udma33 IDE
    to a Promise ata100 lba48 controller card, I never saw the write
    performance go above 1.4MB/s using 100Mbps NICs. The read performance
    was only about 4 times faster at best.

    I gave up on this NOS on account of the 'slow by design' writing
    performance and also on account I couldn't use hdds larger than 120GB in
    spite of all my efforts at finding a solution to the speed and size
    limitation issues. The 'vapourware' claims for no disk capacity limits
    obviously only applied to, as yet, undeveloped and radically different
    future versions of Netware. :-(

    If 'oodles of disk space' is only going to involve drives smaller than
    120GB, ok, otherwise forget it with 160GB and bigger. If the MoBo
    supports LBA48, don't expect faster than 1MB/s writes regardless of
    whether you're using 100 or 1000Mbps NICs. :-(

    HTH

    --
    Regards, John.

    To reply directly, please remove "buttplug" .Mail via the
    "Reply Direct" button and Spam-bots will be rejected.


  3. Re: Hardware opinion: Slow SCSI or fast IDE

    On Sun, 4 Sep 2005 16:16:52 +0100, Johnny B Good
    wrote:

    >The message <431aaddf.3077224@news.zen.co.uk>
    >from nospam@please.invalid (AnthonyL) contains these words:
    >
    >> I need to make a short term 3.12 (yes, that's not a typo) server.

    >
    >> The existing box is a 486 66mhz 128mb ram with a 2Gb SCSI drive and
    >> 10mbit card serving around 8 users. The box only has 16bit slots.

    >
    >> I could try to replicate the above or more easily grab a spare PIII
    >> 350mhz, 256mb RAM 100mbit PCI card with oodles of disk space on the
    >> IDE drive.

    >
    >> Although I've always adhered to using SCSI drives for servers I
    >> suspect that the faster IDE drive with extra cache memory available
    >> will run rings around the older 486 machine.

    >
    >> What do you reckon?

    >
    > Unlikely to make a blind bit of difference. My first experience of
    >netware was 2.15C on an 8MHz 286 box (NEC PowerMate 2, as it happened),
    >with 600MB ESDI disk.
    >


    Stumbled on my 2.12 (Acer OEM) disks the other day and recall
    discussions about ESDI or SCSI. Our first was ESDI also

    > I upgraded the MoBo to a 386SX33 one and saw very little improvement.
    >Having upgraded to a 386 cpu based system, I was then able to ditch the
    >2.15C and install 3.11 which I quicly upgraded to 3.12 which I was using
    >up until a few weeks ago when I replaced it with a Debian based samba
    >server.
    >


    I'd agree that the processor would not make much difference as it was
    not the limiting factor in file and print sharing.

    > After installing 3.12, I regularly upgraded the hardware through 486,
    >SKT7 P166, SKT370 Celeron 466 and (briefly) skt370 p3/500 and finally
    >back down to an under volted and underclocked skt7 K6/500 (running at
    >250MHz) reducing the power consumption to a mere 29 watts with a single
    >250GB Seagate drive.
    >
    > In all that time and in spite of upgrading from the on-board udma33 IDE
    >to a Promise ata100 lba48 controller card, I never saw the write
    >performance go above 1.4MB/s using 100Mbps NICs. The read performance
    >was only about 4 times faster at best.


    You don't seem ever to have gone SCSI. How many users?

    >
    > I gave up on this NOS on account of the 'slow by design' writing
    >performance and also on account I couldn't use hdds larger than 120GB in
    >spite of all my efforts at finding a solution to the speed and size
    >limitation issues. The 'vapourware' claims for no disk capacity limits
    >obviously only applied to, as yet, undeveloped and radically different
    >future versions of Netware. :-(
    >


    I though Netware 5 could handle terabytes -
    http://support.novell.com/techcenter...d19990102.html


    > If 'oodles of disk space' is only going to involve drives smaller than
    >120GB, ok,


    oodles relative to the existing 2gb drive like I mean 6gB


    --
    AnthonyL

  4. Re: Hardware opinion: Slow SCSI or fast IDE

    The message <431c2ed7.11852683@news.zen.co.uk>
    from nospam@please.invalid (AnthonyL) contains these words:

    > On Sun, 4 Sep 2005 16:16:52 +0100, Johnny B Good
    > wrote:


    > >The message <431aaddf.3077224@news.zen.co.uk>
    > >from nospam@please.invalid (AnthonyL) contains these words:
    > >
    > >> I need to make a short term 3.12 (yes, that's not a typo) server.

    > >
    > >> The existing box is a 486 66mhz 128mb ram with a 2Gb SCSI drive and
    > >> 10mbit card serving around 8 users. The box only has 16bit slots.

    > >
    > >> I could try to replicate the above or more easily grab a spare PIII
    > >> 350mhz, 256mb RAM 100mbit PCI card with oodles of disk space on the
    > >> IDE drive.

    > >
    > >> Although I've always adhered to using SCSI drives for servers I
    > >> suspect that the faster IDE drive with extra cache memory available
    > >> will run rings around the older 486 machine.

    > >
    > >> What do you reckon?

    > >
    > > Unlikely to make a blind bit of difference. My first experience of
    > >netware was 2.15C on an 8MHz 286 box (NEC PowerMate 2, as it happened),
    > >with 600MB ESDI disk.
    > >


    > Stumbled on my 2.12 (Acer OEM) disks the other day and recall
    > discussions about ESDI or SCSI. Our first was ESDI also


    > > I upgraded the MoBo to a 386SX33 one and saw very little improvement.
    > >Having upgraded to a 386 cpu based system, I was then able to ditch the
    > >2.15C and install 3.11 which I quicly upgraded to 3.12 which I was using
    > >up until a few weeks ago when I replaced it with a Debian based samba
    > >server.
    > >


    > I'd agree that the processor would not make much difference as it was
    > not the limiting factor in file and print sharing.


    > > After installing 3.12, I regularly upgraded the hardware through 486,
    > >SKT7 P166, SKT370 Celeron 466 and (briefly) skt370 p3/500 and finally
    > >back down to an under volted and underclocked skt7 K6/500 (running at
    > >250MHz) reducing the power consumption to a mere 29 watts with a single
    > >250GB Seagate drive.
    > >
    > > In all that time and in spite of upgrading from the on-board udma33 IDE
    > >to a Promise ata100 lba48 controller card, I never saw the write
    > >performance go above 1.4MB/s using 100Mbps NICs. The read performance
    > >was only about 4 times faster at best.


    > You don't seem ever to have gone SCSI. How many users?


    Four maximum, it was a "Home Network" setup to save me having to
    upgrade the hand me down PCs' hard disk drives my kids were using (Ah,
    the joys of networked Doom! :-)

    SCSI never impressed me performancewise compared to their cheaper IDE
    cousins. In fact (for desktop PC use) they were rather a let down
    performancewise size for size. There was also the issue of heat and
    noise (and excessive power consumption) to consider. For my needs, SCSI
    just didn't cut the mustard.

    > >
    > > I gave up on this NOS on account of the 'slow by design' writing
    > >performance and also on account I couldn't use hdds larger than 120GB in
    > >spite of all my efforts at finding a solution to the speed and size
    > >limitation issues. The 'vapourware' claims for no disk capacity limits
    > >obviously only applied to, as yet, undeveloped and radically different
    > >future versions of Netware. :-(
    > >


    > I though Netware 5 could handle terabytes -
    > http://support.novell.com/techcenter...d19990102.html


    I don't doubt this is true, after all, 3.11 predated win95 by a good 2
    years and was obviously able to support drive capacities well in excess
    of what was then currently available as opposed to dos which kept
    trailing behind these capacity limit increases.

    You've got to hand it to Novell for creating a NOS that only suffered
    this fate after 10 years of hard disk drive development.

    > > If 'oodles of disk space' is only going to involve drives smaller than
    > >120GB, ok,


    > oodles relative to the existing 2gb drive like I mean 6gB


    As I said, I don't think you'll see any improvements then. The
    bottleneck _isn't_ the disk drive. Of course, I never tried it with any
    fancy SCSI HBAs and noisy SCSI drives so I can only comment on IDE
    types. It just struck me as rather strange that despite 'modern' (FSVO
    modern) ATA100 controllers and Novell drivers to match, the 40MB/s or
    faster write speeds of 80 and 120 GB drives was all for naught. :-(

    If I could get my hands on a very cheap (say 5 user) version of Netware
    5, I'd install it in a shot. The SAMBA server only replaces that old
    Netware 3.12 setup on account it supports drives bigger than 120GB and
    brings the writing speed over the 100Mbps LAN closer to the 8.25MB/s I
    usually see win2k box to win2k box. The Debian based server gives me
    about 7.8MB/s write speeds which is over 5 times faster than I was
    previously getting.

    The only downside I've noticed is the slower (most definitely uncached)
    reading of directory trees from the server. I have a CD's worth of mod
    files (3600 odd tunes in some 20 odd directories) stored on the server
    and whenever I load ModPlugPlayer with the full playlist, it takes about
    15 seconds to load B4 it starts playing as opposed to the 4 seconds it
    used to take with the Novell server setup.

    Considering that Linux is free and Novell netware costs hundreds of
    pounds, that's a downside I can live with (and possibly fix :-).

    --
    Regards, John.

    To reply directly, please remove "buttplug" .Mail via the
    "Reply Direct" button and Spam-bots will be rejected.


  5. Re: Hardware opinion: Slow SCSI or fast IDE

    On Mon, 5 Sep 2005 19:01:51 +0100, Johnny B Good
    wrote:

    >The message <431c2ed7.11852683@news.zen.co.uk>
    >from nospam@please.invalid (AnthonyL) contains these words:
    >


    >> You don't seem ever to have gone SCSI. How many users?

    >
    > Four maximum, it was a "Home Network" setup to save me having to
    >upgrade the hand me down PCs' hard disk drives my kids were using (Ah,
    >the joys of networked Doom! :-)
    >
    > SCSI never impressed me performancewise compared to their cheaper IDE
    >cousins. In fact (for desktop PC use) they were rather a let down
    >performancewise size for size. There was also the issue of heat and
    >noise (and excessive power consumption) to consider. For my needs, SCSI
    >just didn't cut the mustard.
    >


    In "single user" type mode it wouldn't impress. SCSI only scores when
    there is concurrent multi-user access and Novell also then scores
    because of its excellent disk caching. Hence my original question.

    Cheers


    --
    AnthonyL

  6. Re: Hardware opinion: Slow SCSI or fast IDE

    The message <431d6872.92126581@news.zen.co.uk>
    from nospam@please.invalid (AnthonyL) contains these words:

    > On Mon, 5 Sep 2005 19:01:51 +0100, Johnny B Good
    > wrote:


    > >The message <431c2ed7.11852683@news.zen.co.uk>
    > >from nospam@please.invalid (AnthonyL) contains these words:
    > >


    > >> You don't seem ever to have gone SCSI. How many users?

    > >
    > > Four maximum, it was a "Home Network" setup to save me having to
    > >upgrade the hand me down PCs' hard disk drives my kids were using (Ah,
    > >the joys of networked Doom! :-)
    > >
    > > SCSI never impressed me performancewise compared to their cheaper IDE
    > >cousins. In fact (for desktop PC use) they were rather a let down
    > >performancewise size for size. There was also the issue of heat and
    > >noise (and excessive power consumption) to consider. For my needs, SCSI
    > >just didn't cut the mustard.
    > >


    > In "single user" type mode it wouldn't impress. SCSI only scores when
    > there is concurrent multi-user access and Novell also then scores
    > because of its excellent disk caching. Hence my original question.


    If you're considering setting up with a SCSI HBA and a 6GB SCSI drive,
    you might discover the secret to getting sustained data transfer rates
    that can match the drive's capabilities, i.e. 2 or more times faster
    than the poxy 1 to 1.4MB/s I was seeing with drives perfectly capable of
    30 to 40 MB/s. The performance hit over a 100Mbps network during read
    operations wasn't as crippled (I was seeing around 4 to 6 MB/s
    transfers).

    Actually, now I think about it, when I had 512MB of ram, I was seeing
    similar 4MB/s writing rates over the wire until the disk writing cache
    filled up (after about 400MB's worth of data had gone over). What got me
    was the enforced 1MB/s write speed to disks quite capable of 30 to
    40MB/s sustained write speeds (using on-board udma33/66 IDE ports, the
    promise ata100 controller only boosted things by a mere 40%).

    I had the latest novell drivers for the relevant chipsets involved (and
    I got to try quite a few!), but there seemed to be a hard limit at 1MB/s
    (1.4MB/s in the promise controller case) which seemed to have absolutely
    nothing to do with disk drive performance.

    If you're going to set up a SCSI based system, I'd be interested in
    your experience, assuming you're going to be using CAT5 at 100Mbps.

    --
    Regards, John.

    To reply directly, please remove "buttplug" .Mail via the
    "Reply Direct" button and Spam-bots will be rejected.


  7. Re: Hardware opinion: Slow SCSI or fast IDE

    The message <431d6872.92126581@news.zen.co.uk>
    from nospam@please.invalid (AnthonyL) contains these words:

    > On Mon, 5 Sep 2005 19:01:51 +0100, Johnny B Good
    > wrote:


    > >The message <431c2ed7.11852683@news.zen.co.uk>
    > >from nospam@please.invalid (AnthonyL) contains these words:
    > >


    > >> You don't seem ever to have gone SCSI. How many users?

    > >
    > > Four maximum, it was a "Home Network" setup to save me having to
    > >upgrade the hand me down PCs' hard disk drives my kids were using (Ah,
    > >the joys of networked Doom! :-)
    > >
    > > SCSI never impressed me performancewise compared to their cheaper IDE
    > >cousins. In fact (for desktop PC use) they were rather a let down
    > >performancewise size for size. There was also the issue of heat and
    > >noise (and excessive power consumption) to consider. For my needs, SCSI
    > >just didn't cut the mustard.
    > >


    > In "single user" type mode it wouldn't impress. SCSI only scores when
    > there is concurrent multi-user access and Novell also then scores
    > because of its excellent disk caching. Hence my original question.


    Second follow up:

    In short, I can only comment on my experience which never involved scsi
    drives in the netware box. It seemed to me that there was a hard limit
    of 1MB/s (1.4MB/s in the case of the promise controller) when writing to
    the disk regardless of the fact that the drives could quite easily
    perform 30 or more times faster than this.

    I know this was to do with the actual disk writes since I could see
    much faster data transfers over the wire until the disk write cache
    buffer filled up, whereupon the rate would drop from the 4MB/s or so
    speed down to 1MB/s (or 1.4MB/s in the promise controller case).

    When you do set up that server (with scsi drive), I'd be interested to
    know if using scsi gets around the 1MB/s write speed limit.

    --
    Regards, John.

    To reply directly, please remove "buttplug" .Mail via the
    "Reply Direct" button and Spam-bots will be rejected.


+ Reply to Thread