• Messages sorted by: [ date ][ thread ][ subject ][ author ]
• [ Next ]In reply to: Jonathan Shapiro

At 11:04 PM -0800 11/17/97, Jonathan Shapiro wrote:
>I am writing (in my thesis) a description of the overhead that goes with the
>combination of externalizing page fault handling (in general) and
>additionally requiring that all pages be backed.
>
>My question is this: was any attempt made to consider a cacheing design in
>the KeyKOS kernel in which the space bank might be in cahoots with the
>kernel about a cache of available storage whose state was included in the
>checkpoint? This would substantially reduce the overhead of "unbacked"
>(i.e. heap/stack) page faults, but I think it's probably complex.
>
>I've no intention of doing this, but I'm curious if it was contemplated.

If I understand correctly, you are looking for a class of pages which are automatically rescinded on restart. We never looked at those. I'm not sure just how much less costly they would be.

On the issues of persistence. The two extremes of:

(1) Save everything (ala KeyKOS), and
(2) Save nothing (e.g. a communication controller)

are easy to reason about and code for. The other variations:

(3) Save some of the stuff (e.g. the word processor that doesn't save your place in the document), and
(4) Save some stuff more frequently (e.g. KeyKOS transaction processing)

are more difficult to reason about and get right.

Bill Frantz       | One party wants to control | Periwinkle -- Consulting
(408)356-8506     | what you do in the bedroom,| 16345 Englewood Ave.
frantz@netcom.com | the other in the boardroom.| Los Gatos, CA 95032, USA

• [ Next ] In reply to: Jonathan Shapiro