Re: performance of rootdbs in non-raw file

I think that there are situations in which using filesystem instead of raw partition is the only solution: raw partitions are not extensible easily on most platforms, except IBM AiX, and some security features of filesystems do not exist in raw partitions. Take IBM AiX for example: JFS already contains a logging system which ensures that recoverable data is ensured. Multiprocessor systems or multihost systems with shared disks (like IBM's HACMP system for high availability, using redondant servers sharing the same disk spaces), often behave much better and are optimized for filesystems, not for raw partitions: these systems know about the underlying filesystem structures and behave consequently, that's what they were built for: ensuring high availability of files, should they be moved transparently from one disk partition to another. SaTriGuy <satriguy@aol.com> a 'crit dans l'article <19970629223200.SAA01895@ladder02.news.aol.com>... > >I have a client who created their rootdbs in a Unix file instead of a > >raw device. I'm wondering what kind of hit this is going to be > >performance-wise. > > I'm not too concerned about this from a performance situation, but am > extreamly concerned about this from a recoverability standpoint.. > > The reserved pages are the most important stuff in the rootdbs and is > extreamly critical for the recoverability of the system. If some of this > is un-recoverable because it is in the UNIX file system buffers at the > point of a system (power) failure, it is possible that the system will not > recover. > Madison Pruet >