Enhance JFS2 defragmentation facility for files and filesystems

Enhance JFS2 defragmentation facility for files and filesystems: alter files in-place, online, to make them contiguous on the filesystem

Inspired by PMR 44659,7TD and other cases from this customer.
Affected applications are writing randomly to large mmap/shmat files, or have a highly random write pattern. When the files are highly fragmented, a relationship is reproducibly established between these write patterns and (a) high IOPS commit cycles and (b) iodone + page-fault-end-processing interrupt waves that grossly affect low-latency application component operation. When the fragmentation is high, the JFS2 IO Gather step is unable to coalesce blocks, and the I/O process becomes inefficient.

Concerted efforts to improve this include reprogramming applications to pre-allocate and pre-initialize blocks using posix_fallocate and write across the file blocks.

There are other variants of this: involving sparse files (PMR 50192,7TD), files with blocks allocated but not initialized, unlinking highly fragmented files (PMR 93866,7TD / DCR #MR0915141358).

In addition to application retooling (and there are many of those), a faster method is needed from the SysAdmin side, to provide relief in Production situations.

Request an enhancement to be able to alter a single file, or multiple files in a filesystem, while in use, to make the files as contiguous as possible. (If Norton Utilities can do it, IBM can too.)