AIX on JS21 Blades - Aix

This is a discussion on AIX on JS21 Blades - Aix ; Hello AIX denizens, I'm faced with a mirroring choice for rootvg on a JS21 blade in an H chassis. The blade has two internal drives and an onboard RAID controller. I plan to use the two internal drives for rootvg ...

+ Reply to Thread
Results 1 to 7 of 7

Thread: AIX on JS21 Blades

  1. AIX on JS21 Blades

    Hello AIX denizens,

    I'm faced with a mirroring choice for rootvg on a JS21 blade in an H
    chassis.

    The blade has two internal drives and an onboard RAID controller. I plan to
    use the two internal drives for rootvg only (user data on a SAN).

    What is best practice for mirroring with JS21s? Let the onboard RAID
    controller handle it (presents one hdisk to AIX), or disable hardware
    mirroring and mirror with AIX rootvg mirroring, as with typical (non-blade)
    servers.

    Thanks,

    Larry Anta



  2. Re: AIX on JS21 Blades

    Larry Anta wrote:
    > Hello AIX denizens,
    >
    > I'm faced with a mirroring choice for rootvg on a JS21 blade in an H
    > chassis.
    >
    > The blade has two internal drives and an onboard RAID controller. I plan to
    > use the two internal drives for rootvg only (user data on a SAN).


    This seems unlikely - the second internal drive precludes the FC
    adapter.

    > What is best practice for mirroring with JS21s? Let the onboard RAID
    > controller handle it (presents one hdisk to AIX), or disable hardware
    > mirroring and mirror with AIX rootvg mirroring, as with typical (non-blade)
    > servers.


    Since you have no option to mirror across PCI busses or PCI adapters on
    a JS21, it would seem a moot point.


  3. Re: AIX on JS21 Blades

    On Wed, 10 Jan 2007 18:38:44 -0800, sjm wrote:

    > Larry Anta wrote:
    >> Hello AIX denizens,
    >>
    >> I'm faced with a mirroring choice for rootvg on a JS21 blade in an H
    >> chassis.
    >>
    >> The blade has two internal drives and an onboard RAID controller. I plan to
    >> use the two internal drives for rootvg only (user data on a SAN).

    >
    > This seems unlikely - the second internal drive precludes the FC
    > adapter.


    Unless an announcement has recently been made that I am unaware of, SAN
    access is _not_ supported on a JS21 running AIX. It will work, because I
    have done it, but it is not supported.

    >
    >> What is best practice for mirroring with JS21s? Let the onboard RAID
    >> controller handle it (presents one hdisk to AIX), or disable hardware
    >> mirroring and mirror with AIX rootvg mirroring, as with typical
    >> (non-blade) servers.

    >
    > Since you have no option to mirror across PCI busses or PCI adapters on
    > a JS21, it would seem a moot point.



  4. Re: AIX on JS21 Blades

    Okay, I'm getting more confused. But please bear with me because I'm
    new (two days experience) to this whole chassis/blade world (although I have
    years of experience with IBM RS/6000s and with AIX in general).

    The H Chassis that we're testing (BladeCenter H Type 8852) has a 20-port
    Brocade FC switch (sorry, no p/n or model handy). This switch *part of the
    chassis*. Six of the switch ports are exposed at the back of the chassis;
    the other 14 ports are internal to the chassis and each of the 14 physical
    JS21 blades is automatically connected to an internal Brocade port. I
    watched the physical installation of two physical hard disks *plus* an FC
    "adapter" (QLogic 4Gb Fibre Channel Expansion Card) into each of the 14
    blades, so I'm confused that this is deemed to be unlikely. Maybe "adapter"
    is the wrong word (If so, sorry about that.) but it's the piece of hardware
    on the blade that permits this internal connection to the Brocade switch.
    With this (hopefully clearer) description of my environment, it had better
    be the case that IBM supports my JS21s since it's their blades, their
    chassis, their (rebranded) FC switch and their SAN technology (DS 4300).

    In any case, that's not what my original question was about.

    I was polling for best practices surrounding the use of the two internal
    hard disks on each blade and was attempting to decide between hardware
    mirroring of those two internal disks (via the blade's onboard RAID
    controller) versus software mirroring with the AIX O/S.

    Normally, choosing between hardware and software mirroring is a no-brainer;
    I'd go with hardware mirroring for performance. My concern was the manner
    in which drive failures are communicated to the AIX administrator, because,
    after all, what good is hardware mirroring if a drive failure goes
    undetected? I've since been advised (by IBM) that drive failures will be
    reported via errpt. This eliminates my concern, so I'm implementing
    hardware mirroring (which is set up by booting off the regular diagnostics
    CD, by the way). With hardware mirroring in place, the AIX admin sees one
    hdisk, even though there are two physical disks behind the scenes. I
    surmise that there is some regular polling of the RAID controller by AIX to
    enable AIX to report (via errpt) that one of the real physical drives has
    failed or that failure is imminent.

    I now feel the decision is justified, but still of course welcome
    discussion. Cheers!


    Larry Anta
    Ryerson University
    Toronto, Canada



  5. Re: AIX on JS21 Blades

    Larry Anta wrote:

    > Normally, choosing between hardware and software mirroring is a no-brainer;
    > I'd go with hardware mirroring for performance. My concern was the manner
    > in which drive failures are communicated to the AIX administrator, because,
    > after all, what good is hardware mirroring if a drive failure goes
    > undetected? I've since been advised (by IBM) that drive failures will be
    > reported via errpt. This eliminates my concern, so I'm implementing
    > hardware mirroring (which is set up by booting off the regular diagnostics
    > CD, by the way). With hardware mirroring in place, the AIX admin sees one
    > hdisk, even though there are two physical disks behind the scenes. I
    > surmise that there is some regular polling of the RAID controller by AIX to
    > enable AIX to report (via errpt) that one of the real physical drives has
    > failed or that failure is imminent.
    >


    No experience with any of the things you are doing, just some common
    sense advice. Kill off a drive somehow while the machine is running and
    see what actually happens so you aren't making assumptions.

  6. Re: AIX on JS21 Blades


    Larry Anta wrote:

    > Okay, I'm getting more confused. But please bear with me because I'm
    > new (two days experience) to this whole chassis/blade world (although I have
    > years of experience with IBM RS/6000s and with AIX in general).
    >
    > The H Chassis that we're testing (BladeCenter H Type 8852) has a 20-port
    > Brocade FC switch (sorry, no p/n or model handy). This switch *part of the
    > chassis*. Six of the switch ports are exposed at the back of the chassis;
    > the other 14 ports are internal to the chassis and each of the 14 physical
    > JS21 blades is automatically connected to an internal Brocade port. I
    > watched the physical installation of two physical hard disks *plus* an FC
    > "adapter" (QLogic 4Gb Fibre Channel Expansion Card) into each of the 14
    > blades, so I'm confused that this is deemed to be unlikely. Maybe "adapter"
    > is the wrong word (If so, sorry about that.) but it's the piece of hardware
    > on the blade that permits this internal connection to the Brocade switch.
    > With this (hopefully clearer) description of my environment, it had better
    > be the case that IBM supports my JS21s since it's their blades, their
    > chassis, their (rebranded) FC switch and their SAN technology (DS 4300).
    >
    > In any case, that's not what my original question was about.


    My mistake - JS21 supports 2 SATA drives AND a Small Form Factor (SFF)
    PCI card at once. The new Qlogic FC adapter (26R0890) is SFF.


  7. Re: AIX on JS21 Blades

    On Thu, 11 Jan 2007 14:49:24 -0500, Larry Anta wrote:

    > Okay, I'm getting more confused. But please bear with me because I'm
    > new (two days experience) to this whole chassis/blade world (although I have
    > years of experience with IBM RS/6000s and with AIX in general).
    >
    > The H Chassis that we're testing (BladeCenter H Type 8852) has a 20-port
    > Brocade FC switch (sorry, no p/n or model handy). This switch *part of the
    > chassis*. Six of the switch ports are exposed at the back of the chassis;
    > the other 14 ports are internal to the chassis and each of the 14 physical
    > JS21 blades is automatically connected to an internal Brocade port. I
    > watched the physical installation of two physical hard disks *plus* an FC
    > "adapter" (QLogic 4Gb Fibre Channel Expansion Card) into each of the 14
    > blades, so I'm confused that this is deemed to be unlikely. Maybe "adapter"
    > is the wrong word (If so, sorry about that.) but it's the piece of hardware
    > on the blade that permits this internal connection to the Brocade switch.
    > With this (hopefully clearer) description of my environment, it had better
    > be the case that IBM supports my JS21s since it's their blades, their
    > chassis, their (rebranded) FC switch and their SAN technology (DS 4300).


    The blades have an HBA and two drives. The switch config as you
    describe above is correct. The problem lies in support. AIX running on a
    JS21 accessing SAN storage via the above is not supported. Linux running
    on the JS21 _is_.

    >
    > In any case, that's not what my original question was about.
    >
    > I was polling for best practices surrounding the use of the two internal
    > hard disks on each blade and was attempting to decide between hardware
    > mirroring of those two internal disks (via the blade's onboard RAID
    > controller) versus software mirroring with the AIX O/S.
    >
    > Normally, choosing between hardware and software mirroring is a no-brainer;
    > I'd go with hardware mirroring for performance. My concern was the manner
    > in which drive failures are communicated to the AIX administrator, because,
    > after all, what good is hardware mirroring if a drive failure goes
    > undetected? I've since been advised (by IBM) that drive failures will be
    > reported via errpt. This eliminates my concern, so I'm implementing
    > hardware mirroring (which is set up by booting off the regular diagnostics
    > CD, by the way). With hardware mirroring in place, the AIX admin sees one
    > hdisk, even though there are two physical disks behind the scenes. I
    > surmise that there is some regular polling of the RAID controller by AIX to
    > enable AIX to report (via errpt) that one of the real physical drives has
    > failed or that failure is imminent.
    >
    > I now feel the decision is justified, but still of course welcome
    > discussion. Cheers!


    I would test to make sure. Pull a drive and see if you get an errpt
    message. I would not assume that it works, because it has been my
    experience that AIX is being treated somewhat as a stepchild on the JS
    series.

    We don't do H/W mirroring. We do the standard AIX config, with hdisk0
    and hdisk1 for each PV, and so far have found it to be acceptable. We
    don't, however, do heavy disk I/O on our blades; they are app servers
    which exist primarily to do network I/O.

    >
    >
    > Larry Anta
    > Ryerson University
    > Toronto, Canada



+ Reply to Thread