D11? (fwd)

Kragen Sitaker kragen@pobox.com
Thu, 5 Aug 1999 15:12:05 -0400 (EDT)


This paper has some really cool ideas on making X faster.

(Dammit, my VHDL-book review, my EOPL review, and my ringworld stuff
still haven't got a chance to go to kragen-tol.  I'll send them soon.
:)

-- 
<kragen@pobox.com>       Kragen Sitaker     <http://www.pobox.com/~kragen/>
Thu Aug 05 1999
94 days until the Internet stock bubble bursts on Monday, 1999-11-08.
<URL:http://www.pobox.com/~kragen/bubble.html>

---------- Forwarded message ----------
Date: Thu, 5 Aug 1999 15:09:03 -0400 (EDT)
From: Kragen <kragen@kirk.dnaco.net>
To: mjk@sgi.com
Subject: D11?

slashdot.org carried a story about the future of X, which drew hundreds
of comments; someone pointed to your early-1996 D11 paper at
http://reality.sgi.com/opengl/d11/.

It sounds like an astonishingly good idea.  Has anyone ever built a
prototype?  Has anyone filed a patent?

(I hope this mail is not part of a thundering herd of Slashdot mail.)

What happens if a process that happens to have pages mapped in the
active context's address space tries to attach to the active context?

It seems that crashing an APP could be much worse than crashing an X
client, but possibly much better than crashing the X server.

How much faster do you expect client<->APP switching to be than regular
process context switches, on CPUs that don't support ASIDs?

I think modifying the address mappings can be significantly slower than
packing and unpacking an X protocol request, in some cases.

Putting the graphics stuff at user-level might improve reliability, but
on PCs where typical graphics cards need their drivers to run with high
I/O priority levels, it won't improve security.  The X server is
trusted completely, because it can use the graphics hardware to
overwrite random areas of physical memory.

-- 
<kragen@pobox.com>       Kragen Sitaker     <http://www.pobox.com/~kragen/>
Thu Aug 05 1999
94 days until the Internet stock bubble bursts on Monday, 1999-11-08.
<URL:http://www.pobox.com/~kragen/bubble.html>