Re: An idea and proposal about system time

[Thomas Schwinge]

Hello!

On Sat, Feb 17, 2007 at 07:39:22AM +0100, olafBuddenhagen@gmx.net wrote:
> On Fri, Feb 16, 2007 at 02:44:46PM +0100, Thomas Schwinge wrote:
> > An open issue to me so far is if we can find an efficient way to
> > continue providing `libshouldbeinlibc''s `maptime' interface.
> 
> The real question to ask is: Why did the original Mach designers pay so
> much attention to an efficient way of getting the system time -- and are
> their reasons relevant to us?
> 
> If the answer is that they aren't, I guess your proposal is fine, and
> maptime() can really be dropped alltogether. If the the anser is that
> they might be, your proposal could be problematic.

In fact had I already tried to measure this: by `syslog'ging each time
`maptime_read' is used.  But for that to yield usable results, first the
`syslog' bug (see my other email) needs to be fixed...  :-) (Also, a
thusly equipped `ext2fs.static' made the system boot hang when ``cleaning
up temporary files''.)

Looking at the source code, I see that it is used inside `pfinet' for
timing issues and a few times in the file system servers.


> (I guess you can map the memory region from the server just as well,
> but time updates might be less reliable when done in a server instead
> of the kernel...)

Yes, that's explicitly what I wanted to avoid.  Otherwise we would have
to guarantee that the time server is considered regularely by the
scheduler -- and that mostly for nothing.


Regards,
 Thomas

signature.asc
Description: Digital signature