Re: KAIO

Mark wrote: > Thanks. In the past I remember seeing recommendations against using > cooked simply because it was not considered 'safe'. The argument was > something along the lines of needing to be sure that data was actually > written to disk rather than stuck in the cache when a system failure > occurs. Safety has never been one of the real issues about using RAW versus COOKED for IDS. It was/is an issue for SE which cannot use RAW disk anyway. IDS ALWAYS writes to COOKED chunks in O_SYNC mode which forces the write to complete to disk before the write system call returns to IDS, so that if IDS thinks the data's on disk, then it is, RAW or COOKED (unless your disk array has writeback cache which ignores the IO O_SYNC signal, but that affects bot RAW and COOKED writes). <SNIP> Art S. Kagel sending to informix-list