On Tue, Jan 08, Wichmann, Mats D wrote: > I may be reading this the wrong way, but: > > > I have some more fails, which are canditates for a waiver: > > > > /tset/LSB.os/aio/aio_error/T.aio_error 1 FAIL > > /tset/LSB.os/aio/aio_read/T.aio_read 1 FAIL > > /tset/LSB.os/aio/aio_read/T.aio_read 11 FAIL > > /tset/LSB.os/aio/aio_return/T.aio_return 1 FAIL > > /tset/LSB.os/aio/aio_write/T.aio_write 1 FAIL > > /tset/LSB.os/aio/aio_write/T.aio_write 12 FAIL > > > > I think all aio test cases should be waiver, it depends heavy if you > > have a SMP or non-SMP system, how fast it is and which other processes > > are running ... > > > This sounds like the behavior of async I/O is unpredictable in > a bad way (influenced by system config and/or load), and it's > proposed that that behavior should be waived? Reading it that > way, anyway, I disagree.
The complete current aio implementation on Linux does not work reliable and we should waive it complete. There is a project with a new aio implementation which should work, but it will need some time. Thorsten -- Thorsten Kukuk http://www.suse.de/~kukuk/ [EMAIL PROTECTED] SuSE GmbH Deutschherrenstr. 15-19 D-90429 Nuernberg -------------------------------------------------------------------- Key fingerprint = A368 676B 5E1B 3E46 CFCE 2D97 F8FD 4E23 56C6 FB4B
