Re: 9.40 buggy!

> > > 3) IO is scheduled on a chunk basis. There is loss of scheduling > > optimization if systems are not configured so that there is a mapping > > between the logical device and the physical device. > > Er .... er ... am I alone in working with particularly technologically > advanced sites? Almost every customer I have now is using some sort of SAN. > One of the characteristics of these SANs is that, such is the level of > indirection between logical devices and physical disks, it is not viable to > make *any* inferences about the physical location of your data. There are two reasons to group IO on a chunk basis. 1) is the old business of trying to take advantage of the physical placement. 2) is to be able to buddy-bunch multiple small IO blocks into a single larger IO request. Even with SAN environments, the second reason is still valid. By sorting the IO writes during a checkpoint, we are able to use large buffer writes, and thus reduce the total number of physical writes that are required.