Organization of the file system
Different algorithms are used to manage the cache, but the most common, with
requests to verify whether reading corresponding block buffer cache, and if not,
at first considered his clipboard. Neat caching implementation requires several
problems. First, the limited size of buffer cache. When the unit must be loaded
in the buffer cache filled, there is a problem of replacement units, ie separate
blocks should be removed from it. This situation is very much like the situation
with vytalkivaniem pages of memory, and here uses the same algorithms, such as
FIFO, Second Chance and LRU. The only difference is that cache links are not so
frequent as links to memory. Replacement units should take into account their
importance to the file system. Boxes should be divided into categories, such as:
Blocks index nodes, blocks indirect addressing, blocks directories filled with
blocks of data, etc., and depending on the accessories unit to a category
applied to the different substitution strategy. Secondly, because caching
mechanism uses a deferred account, in which the buffer is changed immediately
write to disk, a major problem is the aging of the information in the disk
blocks, which are images in the buffer cache. Late synchronization and disk
buffer cache may lead to a very undesirable situation, in the case of equipment
failures or software. Therefore, the strategy and the order to display
information from the cache to disk must be carefully designed. Thus, significant
blocks for the file system compatibility (Blocks index nodes, blocks indirect
addressing, directories units), should be rewritten to disk immediately,
regardless of what part of LRU chain they are. There is a need to carefully
choose the order of this rewriting. In UNIX is available for this challenge SYNC,
which leads all modified blocks immediately written to disk. To synchronize the
contents of the cache periodically, and the disk gets protsess-demon background,
which makes it after a certain period of time (for example, every 30 seconds.).
You can also organize a simultaneous operation with individual files, asked at
the opening of the file where all changes to a file stored on disk immediately.
In OS MS-DOS, unlike Unix, a record of each modified block to disk as soon as
possible. The consequence of such a strategy-removing the floppy from the UNIX
system without summoning SYNC almost always leads to a loss of data, while MS-DOS
there is no problem (in UNIX, it is anticipated that all the hard disks and not
ejected) Caching is not the only way to increase system performance. Another
important technology, the reduction in the number of disk head movements reader
through reasonable accommodation strategy information. For example, the array
index of Unix knots trying to place a secondary tracks. Also, it makes sense to
deploy index nodes nearby blocks of data to which they refer, etc.
Next part
|
O
wlCom Software
