[Bug 1204 (linux mm)] New - 2.5?: read-ahead + drop-behind for everything

[bugzilla-daemon@distro.conectiva.com.br]

http://distro.conectiva.com.br/bugzilla/show_bug.cgi?id=1204

--- shadow/1204	Fri Feb  2 17:13:26 2001
+++ shadow/1204.tmp.15345	Fri Feb  2 17:13:26 2001
@@ -0,0 +1,36 @@
++============================================================================+
+| 2.5?: read-ahead + drop-behind for everything                              |
++----------------------------------------------------------------------------+
+|        Bug #: 1204                        Product: Linux kernel MM         |
+|       Status: NEW                         Version: 2.5 (beta)              |
+|   Resolution:                            Platform: All                     |
+|     Severity: enhancement              OS/Version: Linux                   |
+|     Priority: P3                        Component: linux mm                |
++----------------------------------------------------------------------------+
+|  Assigned To: linux-mm-bugs@nl.linux.org                                   |
+|  Reported By: riel@conectiva.com.br                                        |
+|      CC list: Cc:                                                          |
++----------------------------------------------------------------------------+
+|          URL:                                                              |
++============================================================================+
+|                              DESCRIPTION                                   |
+Linux currently has pretty good read-ahead code and pretty good IO clustering
+code; unfortunately the read-ahead code can only be used for read() and the
+clustering code can only be used for mmap() (and, to a certain extent, swap).
+
+It would be better if both the read-ahead code and the clustering code could be
+used for both, as required by the situation. When we have this, it's trivial to
+also honour the madvise flags for every type of memory.
+
+Ben LaHaise has some code available to make read-ahead possible for mmap()ed
+files, AFAIK no code exists yet to do read-ahead for anonymous memory.
+
+For drop-behind for mmap() and anonymous memory no code has been written yet,
+but it should be relatively simple to abuse try_to_swap_out() for this. 
+
+
+Benefits of these improvements would be mostly people with predictable access
+patterns. This should include both the scientific calculation folks and the
+people who regularly switch desktops on low-memory machines (don't we all?).
+Of course the database people will also be happy when Linux is ready to honour
+MADV_RANDOM on any type of memory..
\ No newline at end of file

-
Linux-mm-bugs:  bugzilla list for the Linux-MM subsystem
Archive:        http://mail.nl.linux.org/linux-mm-bugs/
Web site:       http://www.linux.eu.org/Linux-MM/
Development:    linux-mm@kvack.org