Summary: 4 hosts per D1000 - no terminator

From: Eric Forgette (4jet_at_overnite.com)
Date: 05/30/03

  • Next message: Azzopardi Konrad: "Hard disks problem"
    Date: Fri, 30 May 2003 09:10:05 -0400
    To: sun managers <sunmanagers@sunmanagers.org>
    
    

            Well, unfortunately only I had one response, and it was a request to
    clarify my environment. Sun wont budge on this, and strangely enough,
    they wont provide a technical reason why. So my choices are, support
    it myself or remove 1 hba from each scsi bus and terminate the D1000s.
    I don't have any problem supporting it myself, but that leaves my
    employer is a pretty bad position if I get hit by the proverbial bus.
    So I've gone thee Sun supported route, I've terminated the D1000 and
    mirrored the rootdg off into my san. This is kind of a waist of high
    performance disk space, but at least the machine will be up if
    something happens on my single scsi path...

    Issue with my bus sharing configuration described in my original post
    are:

    1.) 1 bad hba may adversely affect a good hba on the same bus (two
    domains or hosts affected by one hardware error).
    2.) Maintenance requires more thought and planning because you may need
    to break a scsi chain.
    3.) Each domain must be booted one at a time because Solaris/OBP does a
    SCSI reset during boot (this panics a booting domain, but not a fully
    loaded and running kernel).

    Since everything is mirrored, I haven't had a problem with any of these
    issues. I used this configuration of over a year without any issues.
    Obviously use at you own risk as Sun wont support you!

    Regards,
    -Eric

    ----
    Eric P. Forgette
    Unix Systems Administrator
    http://homepage.mac.com/e4jet/sysadm/
    Here is the original post:
    Hello,
    An urgent request as always.  I'm looking for the technical reason why 
    Sun
    wont support 4 hosts connected to 2 D1000 (each host having two paths to
    each boot disk).  After reading "SCSI-Initiator ID" By David Deeths (Sun
    BluePrints OnLine - August 2000), and having a pretty good understanding
    of how SCSI busses work, it seems this should work great!  In fact, I've
    been running 4 E10K domains duel pathed to 2 D1000s for over a year
    without any issues (all scsi ports on the back of the D1000 have host 
    buss
    adapters hooked to them).  As long as you're careful to boot each box,
    which shares a bus, one at a time (Solaris/OBP does a SCSI reset during
    boot), it would seem that everything is cool.  I admit that debugging an
    HBA issue would be more challenging, but one could just add a terminator
    and test the suspect HBA.  Does anyone have any technical reasons why 
    this
    isn't supported?  Is anyone doing anything similar to this?
    Regards,
    -Eric
    _______________________________________________
    sunmanagers mailing list
    sunmanagers@sunmanagers.org
    http://www.sunmanagers.org/mailman/listinfo/sunmanagers
    

  • Next message: Azzopardi Konrad: "Hard disks problem"

    Relevant Pages

    • SMP support problem - panic: APIC: Previous IPI is stuck
      ... disabling SMP support in kernel which can not handle the traffic. ... <ACPI PCI bus> on pcib0 ... Waiting 15 seconds for SCSI devices to settle ... device em # Intel PRO/1000 adapter Gigabit Ethernet Card ...
      (freebsd-stable)
    • Re: FreeBSD 5.3 freezes under heavy hdd load
      ... options SOFTUPDATES# Enable FFS soft updates support ... deviceatapicam# allows ATAPI devices to be accessed through SCSI ... devicescbus# SCSI bus ... #deviceem# Intel PRO/1000 adapter Gigabit Ethernet Card ...
      (freebsd-questions)
    • 2.6.15.1: persistent nasty hang in sync_page killing NFS (ne2k-pci / DP83815-related?), i686
      ... the SCSI controller and disks and moving them over to a new machine. ... although md is configured in the new .config, ... the hardware is blameless here. ... # ACPI Support ...
      (Linux-Kernel)
    • [2.6.15 / sata_mv] Oops!
      ... Hardware: 8 SATA drives spread across two MV88SX50x1 controllers built into motherboard. ... 03:01.0 SCSI storage controller: Marvell Technology Group Ltd. ... abnormal status 0xD0 on port 0xF90A211C ... # ACPI Support ...
      (Linux-Kernel)
    • Re: test10, hangs at aic79xx probing
      ... well at the same point without having SCSI devices initialized. ... Linux Plug and Play Support v0.97 Adam Belay ... # Device Drivers ...
      (Linux-Kernel)