Date: Thu, 15 Feb 1996 08:14:42 -0500 (EST) From: "Ron G. Minnich" <rminnich@Sarnoff.COM> To: Alan Cox <alc@cs.rice.edu> Cc: hackers@freebsd.org Subject: basic VM issue Message-ID: <Pine.SUN.3.91.960215081005.14114E-100000@terra> In-Reply-To: <199602150739.BAA16225@noel.cs.rice.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
Alan in his message makes an important point. The Mach VM tends to think of pages as units, rather than thinking of objects as the unit, and so in many places in the xxxbsd vm code you see functions (e.g. vm_fault) which operate on one page at a time. This results in lower-level code having to attempt to re-aggregate page-sized requests into larger units. It is really strange to walk the whole page fault path and see how many times information is lost and then recreated. This is another area where sunos got it right: their vm tends to operate in terms of (offset,length), i.e. in blocks of data, and if fragmentation needs to happen it will happen at the lowest level, not the highest. ron Ron Minnich |" XNFPREP: ERROR 4007: rminnich@sarnoff.com | Everything in the design was deleted." (609)-734-3120 |Was it something I said? ftp://ftp.sarnoff.com/pub/mnfs/www/docs/cluster.html
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SUN.3.91.960215081005.14114E-100000>