DragonFly BSD
DragonFly kernel List (threaded) for 2010-02
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: kernel work week of 3-Feb-2010 HEADS UP

From: Michael Neumann <michaelneuma@xxxxxxxxxxxxxx>
Date: Thu, 4 Feb 2010 02:13:59 +0100

2010/2/4 Matthew Dillon <dillon@apollo.backplane.com>

:Awesome :)
:Would such a SSD cache also improve write performance of a HAMMER file
:To improve write performance we'd probably need to put the UNDO/REDO log
:on a SSD...

   No, the first iteration is strictly a clean slaved throw-away copy.

   Writes run to the actual media and free any related underlying swap
   copy.  A new swap copy may be made later when the cached VM page
   associated with the write cycles back out of main memory.

Understood. So this basically means that when I do a read-after-write I don't
have to read from disk as it's either in memory, or if I wait too long, it's moved
to swap-cache and I still can access it quickly.

:And will the new system also cache mmap'ed data of a file?
:  Michael

   Yes.  Both read() and mmap()'d data are backed by VM pages and are
   applicable to the cache.




[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]