Re: Async I/O at AIX 5.2

From: JOSEPH KREMBLAS (JKREMBLAS_at_REDHEARTGIFTS.COM)
Date: 03/30/04

  • Next message: Ralph R. Rye: "Re: Async I/O at AIX 5.2" 
    Date:         Mon, 29 Mar 2004 15:04:12 -0700
To: aix-l@Princeton.EDU

    The AIX 5.2 Release Notes contains a section that describes AIO at
    http://publib.boulder.ibm.com/pseries/aixgen/relnotes/52RELNOTES/1007390
    3.htm. The text follows.
    -----
    Asynchronous I/O (AIO)

    Legacy AIO has a new subroutine called aio_nwait_timeout
    (_posix_aio_nwait_timeout in POSIX). It extends the capabilities of the
    aio_nwait subroutine by adding a fourth parameter to specify timeout
    values. It is not available through the /usr/include/aio.h file or
    libc.a library, so it must be accessed directly from the kernel.

    The aio_nwait_timeout subroutine is a new system call that is exported
    from the AIO subsystem and is declared as the following:

    int aio_nwait_timeout (max_cnt, req_cnt, aiocbpa, timeout)
    int max_cnt;
    int req_cnt;
    struct aiocbp **aiocbpa;
    uint timeout;
    The first three arguments of this subroutine are the same as the first
    three arguments of the aio_nwait subroutine. The aio_nwait_timeout
    subroutine waits for a certain number of asynchronous I/O operations to
    complete as specified by the req_cnt parameter, or until the call has
    blocked for a certain duration specified by the timeout parameter.

    The max_cnt parameter indicates the maximum number of pointers to the
    aiocbp structure that can be copied into the aiocbpa array.

    The timeout value is specified in units of milliseconds. A timeout value
    of -1 indicates that the subroutine should behave like the aio_nwait
    subroutine, blocking until all of the requested I/O operations complete
    or until there are no more asynchronous I/O requests pending from the
    process.

    A timeout value of 0 indicates that the subroutine will return
    immediately with the current completed number of asynchronous I/O
    requests. All other positive timeout values indicate that the subroutine
    needs to block until either the timeout value is reached or the
    requested number of asynchronous I/O operations complete.

    The aio_nwait_timeout subroutine has to be redefined and accessed from
    POSIX AIO as _posix_aio_nwait_timeout.

    To access the subroutine in Legacy AIO when compiling, use the
    aio_nwait_timeout.imp import file, which contains the following:

    #!/unix

    aio_nwait_timeout
    Assuming the aio_nwait_timout subroutine is called inside of the
    testcase.c file, compile it with the following:

    cc -D_AIO_AIX_SOURCE -o testcase -bI:./aio_nwait_timeout.imp testcase.c

    To access the subroutine in POSIX AIO when compiling, use the
    aio_nwait_timeout.imp import file which contains the following:

    #!/unix

    _posix_aio_nwait_timeout
    Assuming the _posix_aio_nwait_timeout subroutine is called inside of the
    testcase.c file, compile it with the following:

    cc -o testcase -bI:./aio_nwait_timeout.imp testcase.c
    -----

    -----Original Message-----
    From: IBM AIX Discussion List [mailto:aix-l@Princeton.EDU] On Behalf Of
    pSeries AIX Geek
    Sent: Monday, March 29, 2004 2:46 PM
    To: aix-l@Princeton.EDU
    Subject: Re: Async I/O at AIX 5.2

    It's not that complete (as I'd expect from IBM docs).

    Anyone have real-world experience?

    - pAG

    --- Bob.Kelley@BRINKSINC.COM wrote:
    > found this reference but I haven't read up yet
    >
    >
    http://publib16.boulder.ibm.com/pseries/en_US/aixprggd/kernextc/async_io
    _subsys.htm
    >
    >
    >
    > Bobby Kelley Jr.
    > 972-877-5341
    >
    >
    >
    > |---------+---------------------------->
    > | | pSeries AIX Geek |
    > | | <aixgeek@YAHOO.CO|
    > | | M> |
    > | | Sent by: IBM AIX |
    > | | Discussion List |
    > | | <aix-l@Princeton.|
    > | | EDU> |
    > | | |
    > | | |
    > | | 03/28/2004 07:09 |
    > | | PM |
    > | | Please respond to|
    > | | IBM AIX |
    > | | Discussion List |
    > | | |
    > |---------+---------------------------->
    >
    >
    >-----------------------------------------------------------------------
    ----------------------------------------------------------------------|
    > |
    >
    > |
    > | To: aix-l@Princeton.EDU
    >
    > |
    > | cc:
    >
    > |
    > | Subject: Async I/O at AIX 5.2
    >
    > |
    >
    >
    >-----------------------------------------------------------------------
    ----------------------------------------------------------------------|
    >
    >
    >
    >
    > There are two different aio devices at AIX 5.2:
    > aio0
    > (legacy) and posix_aio0 (new at AIX 5.2).
    >
    > Anyone have any comments on the difference?
    >
    > aio0 is marked as the "legacy" version (run "lsdev
    > -C
    > | grep aio"), but yet, it's different, in that
    > maxservers/minservers are set on a CPU basis, not system-wide.
    >
    >
    > - pAG
    >
    > __________________________________
    > Do you Yahoo!?
    > Yahoo! Finance Tax Center - File online. File on
    > time.
    > http://taxes.yahoo.com/filing.html
    >
    >
    >
    >
    > NOTICE TO BRINK'S EMPLOYEES:
    > All electronic mail communications composed, sent,
    > received or stored on or
    > through Brink's equipment are subject to the Brink's
    > Electronic
    > Communications Policy. Employees should have no
    > expectation of privacy in
    > their use of the Company's electronic mail system.

    __________________________________
    Do you Yahoo!?
    Yahoo! Finance Tax Center - File online. File on time.
    http://taxes.yahoo.com/filing.html

  • Next message: Ralph R. Rye: "Re: Async I/O at AIX 5.2"