HP OpenVMS Systems

ask the wizard
Content starts here

I/O Data Loss (Stalls) under High Load?

» close window

The Question is:

I have a DMA device feeding data into the ALPHA at 30 MBytes per sec. non-stop.
  I must transfer that data to a RAID system before more data comes in to
 overwrite what I have.
Currently I use 8 rotating input buffers of 10 MBytes each and use QIO with AST
 routines.  I can write about 15GBytes of data before I somehow get interrupted
 and lose a few miliseconds worth of data. I am running at priority 63.  What
 techniques can I us
e to keep from getting interrupted, or find out what is doing it?  The Alpha is
 completely dedicated to this process and I have complete control over what
 runs on it and when.  When I set up the system using hand-shaking, I can
 sustain data rates of 63 MB
ytes per second without losing data so I know my RAID system and VMS are
 running fast enough.  But when I remove hand-shaking (which will be the
 real-world case) I sometimes lose those chunks of data.

The Answer is :

  Please contact the customer support center, as this could arise from
  anything ranging from a hardware error to a software problem -- many
  potential sources of interruptions exist -- and thus details on the
  RAID system and the particular Alpha will be of interest, as will
  details on the incoming data path and the RAID data path will be of
  interest.  Details on exactly what you mean by hand-shaking will also
  be of interest -- there are many different hand-shaking protocols.
  The OpenVMS Wizard will assume you have ruled out an error within
  the incoming data stream.
  The OpenVMS Wizard will also assume you are using 64-bit addressing.
  Processes using 32-bit addressing are limited to one (P0) to two
  (P0 and P1) gigabytes of address space.  Fast I/O may also be
  of interest, and current or expected V7.3-2 Fast Path features
  (depending on the RAID device) if this is an SMP configuration.
  The OpenVMS Wizard will further assume you have all current ECO
  kits installed for this configuration.
  For information on synchronization bugs and related coding errors,
  please see topic (1661) -- high loads can uncover latent bugs in
  application code and even in OpenVMS itself.  (In the experience of
  the OpenVMS Wizard, the former case is more common than the latter.)
  Again, please contact the support center or services organization,
  as a far more detailed investigation will be required here.

answer written or last revised on ( 19-AUG-2003 )

» close window