PF module option definitions

The pf module is a data prefetching module for the MIO user program and is optional.

nopffw: Do no prefetch pages when in write mode. This is the default option.; The default behavior of the pf cache is to not pre-read a page into the cache if the call that triggered the pre-read is from a user write into the cache. This because there is no benefit to reading in a page just to have it overwritten. But if a subsequent user write into the page is ill-formed (not beginning and ending on sector boundaries), the logic that marks dirty sectors would no longer be valid, so the dirty pages need to be synchronously written to disk and then the entire page is read synchronously into the cache. In this case it is better to just asynchronously read the page to be overwritten, rather than take the synchronous hit on a page miss.

release: Free the global cache pages when the global cache file usage count goes to zero. This is the default option.

norelease: Do not free the global cache pages when the global cache file usage count goes to zero.; The release and norelease options control what happens to a global cache when the file usage count goes to zero. The default behavior is to close and release the global cache. If a global cache is opened and closed multiple times there may be memory fragmentation issues at some point. Using the norelease option keeps the global cache opened and available even if the file usage count goes to zero.

private: Use a private cache. This means that only the file opening the cache may use it.

global: Use a global cache. This means that multiple files can use the same cache space. This is the default option.; It has been our experience that a global cache is most often the best option for several reasons. The amount of memory is deterministic as you will know how many caches are opened and how large they are. Using private caches, one may not know how many private caches are active simultaneously. A maximum of 256 global caches can be opened. The default is to use set the global option to zero, which means that one global cache is open. Each global cache opened can be assigned by the user to a particular group of files.

asynchronous: Use asynchronous calls to the child module. This is the default option.

noasynchronous: Alias for synchronous.; The asynchronous, synchronous, and noasynchronous options control whether the cache should use asynchronous I/O calls to load cache data from the filesystem. Sometimes used for debugging or when aio is not enabled on the system.

nodirect: Do not use DIRECT I/O. This is the default option.; The direct and nodirect options control whether the O_DIRECT bit is OR'd into the oflags on file open. The pf cache is capable of doing direct I/O. The cache aligns all cache pages on 4K boundaries and attempts to issue only well-formed requests that are needed to ensure that I/O requests go direct.

cache_size: The total size, in bytes, of the cache. Sizes in kilobytes, megabytes, gigabytes, and terabytes are also recognized. Default value is 64k.

page_size: The size, in bytes, of each cache page. Sizes in kilobytes, megabytes, gigabytes, and terabytes are also recognized. Default value is 4k.

stats{=output_file}: Output prefetch usage statistics: file name for pf output diagnostics.; If no output_file is specified or if it is mioout, which is the default value, the pf module searches for an output statistic file defined in the MIO_STATS environment variable.

inter: Output intermediate prefetch usage statistics on a kill -SIGUSR1 command.

nointer: Do not output intermediate prefetch usage statistics. This is the default option.; The inter option instructs the pf cache to output usage statistics when a kill -30 command is received by the application.

notain: Do not retain file data after close for subsequent reopen. This is the default option.; The retain option instructs a global pf cache to save the file's pages in the cache to be reused if the file is reopened in the same global cache. The underlying file must not have been modified between the close and open. The file's pages in the cache are still subject to LRU preemption as though the file were still opened in the cache.

nolistio: Do not use listio mechanism. This is the default option.; The cache normally does not use listio, which has been used primarily for debugging.

notag: Do not use prefix stats flow. This is the default option.; The tag string is a prefix for output printed to the stats file. When the stats file gets large this will make it easier to search for sections of interest.

noscratch: The file is to be flushed saved on close. This is the default option.; Instructs the cache to treat a file as scratch, meaning the file is not flushed at close and is unlinked after closing.

passthru: Byte range to pass thru the cache.; The user can specify a range of the file that is not cached. The data is read and written through the cache. This has been necessary in the past for parallel applications that write to the same file and share header information at the start of the file.