Re: AIO vp...

Art S. Kagel wrote: > > Stefan Weideneder wrote: > > > > Art S. Kagel wrote: > > > > > > SUNIL S. SINGH wrote: > > > > > > > > Hi all !!! > > > > > > > > I have a little doubt here abt AIO vp. > > > > One of the informix manual says tht IDS uses kernel AIO (kio) I/o method, > > > > ie performing I/O through operating system, if chunk is UNIX raw device > > > > and it relies on AIO vp for performing I/O directly if chunk is UNIX > > > > cooked file. > > > > > > > > How far is it correct ?? > > > > I think it is other way round. > > > > > > No that is correct. By default, on most platforms, RAW device chunks > > > are handled by KAIO threads and COOKED device/file chunks are handled > > > by AIO VPs. > > > > > > Art S. Kagel > > > > ... on most platforms, except some platforms, i.e. HP-UX. > > Well let's be clear, Stefan. Both our posts leave the question of > exactly what is platform dependent open. The dependency is whether > KAIO is enabled or disabled by default. If KAIO is enabled, either > implicitely (because it is the default and the KAIOOFF environment var > is not defined) or explicitely (because the KAIOON environment var is > defined where KAIO is not the default), then KAIO handles RAW chunks and > AIO VPs handle COOKED chunks. If KAIO is not enabled, either because > it is not implemented for your platform, it has been disabled, or it > has not been enabled (depending on the platform default), then AIO VPs > handle all I/O. > > There that's better. Sorry to have been muddy. > > Art S. Kagel Seems to be version dependent, too. On a few new platforms you can see kaio-threads running in the AIO-vp's. First detected on NT-machines, then I saw this phenomenon on IDS 7.30 UNIX boxes, too. Stefan Weideneder