From: Scott A C. <sc...@cs...> - 2002-08-02 20:13:32
|
On 02 Aug 2002 13:36:23 -0500, Cynbe ru Taren <cy...@mu...> writes: > > The Muq design is 12 years old now: At this point, I'm > inclined to put more stress on programmer convenience and > less on space and time efficiency. With gigabytes of ram > now routine in cheap PCs, does anyone really care about > disk-basing? Much of the coding effort of the Muq I wouldn't worry about diskbasing objects containing references, but diskbasing blobs (binary objects containing no references) might still be considered a win. You get almost all of the simplicity of an exclusively in-RAM GC, and you get most of the storage benefits of diskbasing. If someone has 4gb of stuff, its probably safe to say that 95% of it is going to be in the form of large blobs? Then lightweight proxy objects for each such external object.. Say 16-24 bytes bytes. 4 bytes object-location-id. 4 bytes size. 4 bytes for in-memory location. (null if not in memory). 4 bytes for maintance (track dirtyness, track usage to identify when not used.), maybe some hashbits? > Just for fun, I've also been writing simple C++ hacks to > generate raytraced images of hierarchical cubic spline > surfaces via RenderMan descriptions. Great > instant-gratification programming! Something I don't do > enough of. I can now make absolutely marvelous random blobs. > :) This might also wind up in Muq eventually -- it is based Got pictures? Scott |