Re: [Syllable-kernel] more on EROS and KeyKOS..

[Mark W. <ma...@cl...>]

> On the whole, an interesting idea.  Unfortunately, I believe it dates
> itself.  Memory is way to large to be written to disk every 30 seconds.
 <snip>
> a fast application can already dirty memory faster
> than it can be written to disk with full-time streaming writes
<snip>

It seems to me you could do some tricks to get round these problems...

To me, this resembles the problems involved in live-migrating virtual 
machines.  For this purpose, the Xen (http://xen.sf.net) VMM tracks process' 
page dirtying by setting all the pages read only and recording the faults.  A 
similar mechanism could be used in a checkpointing OS to determine which 
pages changed between checkpoints.

In a paper on the "self-migration" scheme developed by Jacob Gorm Hansen 
(http://www.diku.dk/~jacobg/self-migration/), the possibility of temporarily 
"stunning" processes with high dirtying rates is mentioned - sounds like it 
could also be applied in this sort of system.

> Still, it's an interesting concept.

Yes.  I attended a talk by one of the EROS project leads and apparently the 
orthogonal persistance thing was also an aid to security proofs: (apparently) 
it's much easier to prove that the system will stay secure given a secure 
starting state than to start your argument with system startup, so 
eliminating true reboots makes life easier.

Cheers,
Mark