Re: L4-HURD , POSIX, UNIX

[Jonathan S. Shapiro]

On Mon, 2005-11-07 at 02:50 +0100, address@hidden wrote:

> I think we all agree that this kind of recovery *is* a killer feature.
> But it isn't tied to persistance; there are various mechanisms that
> could achieve this, not at all necesserily as radical.

Antrik:

I find it very interesting that you consistently reject a simple
mechanism that is known to work quite well in practice favor of other
persistence mechanisms that are known to work quite badly in practice.
Implementing persistence on a per-process basis is a nightmare. There
are many toolkits that do it, and it is the source of a great deal of
extra programmer effort. It is certainly *not* something that you can
slip under an existing POSIX interface without changing the application.

These other mechanisms are also dramatically less efficient, and they do
NOT provide the transaction-like properties that are so useful in the
machine-wide mechanism.

I would also like to understand why you characterize the machine-wide
approach as "radical". Can you expand on this?

> Of course, this still leaves quite a lot of choices about the internal
> architecture of the system. However, as for user-visible changes,
> radical concepts like system-wide transparent persistance are definitely
> ruled out IMHO.

Let us test this.

Several years ago, Intel introduced a "save to memory" feature in
laptops. You close the top of your laptop case. Later you reopen it and
all the applications are still running. Do you think that users find
this radical or strange?

And by the way, *that* persistence stops at the machine boundary too.

shap