Re: Cooked files and mmap

Pablo Sanchez wrote: > > In article <5n1if2$8rm@cssun.mathcs.emory.edu>, Denham.M@amstr.com writes: > > > > Are there new features at the O/S level (such as journalling?) that > > ensure that this type of scenario does not become a problem with > > cooked files? > > > > On SGI (please note, tech marketing hat is now on), you can do > directed I/O. The ability to do directed I/O allows you to have user > level async I/O along with block access and having the data on disk > map directly to user space (rather than going to kernel buffers and > bcopy'ing to user space). Take that hat off!!! ;-) The issue of cooked verus raw *will* depend on the platform and hardware. The scenario originally written about was buffered I/O which can occur in either raw or cooked disk partitions. Informix went to turbo to reduce the overhead of UNIX I/O. This was back in '89-90 time frame. Vendors like SGI have enhanced the I/O of their OSs so that they can deal with disk intensive apps. (DB, video on demand, etc...) Raw partitions will in most cases perform better than cooked. Use cooked iff (that if and only if) you intend to place the db in a site where remote management is going to be difficult and there is not a local admin. Again, if this is the case, I'd use SE for reliability..... Just a few stray thoughts from your favorite Uncle... -Mike