Subject: None
To: None <tech-kern@NetBSD.ORG>
From: Chris Csanady <>
List: tech-kern
Date: 03/01/1996 16:01:55
i was curious, how terribly would it bother people if we moved update back
into kernel space?  it just seems that if soft updates were to be implemented,
there would need to be a background daemon in the kernel anyway.  i am really
not quite sure about the best way of implementing soft updates, and i really
dont think im qualified to invent something new.  suggestions are welcome
though. :)

seconly, which should have come first, would the current implementation of
soft updates as described by the paper be acceptible?  i know this would
require adding a pointer to a dependancy structure from withing the buf
struct, and routines to guarantee that the buffers get written back to disk
in the proper order.  in other words, it would not be contained souly in the
fs code. :(  would this be okay though?

i would really like to attempt to implement this, but i also dont want to
waste my time. ;)  i think that increased performance in metadata writes,
as well as overall concurency would be a big win.  i really havent done
much kernel work as of now, but i would really like to give it a try.  if
anyone could point me to some references on the vfs/vnode interfaces, and
perhaps the buffer cache, i would apreciate it.  i think i have a basic
understanding of things now, but i only have time to read so much source. :)