Re: Persistence

[Jonathan S. Shapiro]

On Fri, 2005-10-28 at 07:49 +0200, Martin Schaffner wrote:
> On Thu, 27 Oct 2005 22:08:39 -0400, Jonathan S. Shapiro wrote:
> 
> > As to cost, yes, I am afraid that EROS-style persistence is
> > significantly more efficient than conventional file systems, and we
> > would face a definite design burden finding interesting uses for the
> > newly available disk bandwidth.
> 
> If I understand correctly, then what persistence does is keep a memory 
> dump of everything on disk, so a power outage does not destroy any 
> data. This also makes constant "saving" unnecessary. I guess this 
> mechanism is related to swapping (the app tells the system which pages 
> it will need in the future, the OS pages them in and out, and a power 
> failure just means that the app needs to be paged in again). Is this 
> true?

Yes. This is about right. There is some additional detail to make sure
that the snapshot is properly transacted, but you have the idea. If you
want all of the gory details:

        http://www.eros-os.org/papers/storedesign2002.pdf

> If yes, then when one wants to back up a document or transfer it to 
> another machine, these memory dumps might not be very useful, as 
> traditionally memory dumps are machine-dependent (byte order, 
> alignment, absolute pointers) for speed, while file formats are 
> designed to be machine-independent.

You still need to be able to export documents into some linearized form.

But if you stop to think about it, you will conclude that you needed
this for document interchange as well. If you need to do cut and paste,
the two applications need a reference format that is known to both.

Also, you want to save the document at some point and exit the editor.
This also requires that you serialize the document. So editors still
need a save function.

Where persistence becomes really helpful is (a) in implementing things
like file systems, and (b) in keeping complex arrangements of processes
in sync with each other.


shap