Mailinglist Archive: opensuse (2217 mails)

[Bill Anderson <bill@xxxxxxxxxxxxxxxxx>]

Carlos E. R. wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
>
> The Thursday 2007-09-20 at 09:53 -0600, Bill Anderson wrote:
>
>   
> > > > kernel only needs to send the anonymous memory pages of an 
> > > > application to swap space, the text (code) can be retrieved from the 
> > > > disk file for the application.
> > > >     
> > > >         
> > > Which is probably slower than just using swap for all (instead of seeking
> > > all around the disk).
> > >   
> > >       
> > This paging concept has been around for awhile. Even Unix has doesn't page
> > text to swap, as the writes are just too expensive. With a good elevator
> > algorithm the cost of seeks is minimal.
> >     
>
> I know, even windows uses that method. But the real reason was that swap 
> space was expensive, when the method was invented; that is no longer the 
> case. 
>
> There is a noticeable difference in speed from reading code from each 
> respective file than compared to read from contiguous raw swap space.
>
> It can be worse if it first have to read the inode, then the code, which I 
> think it might do.
>
> - -- 
> Cheers,
>        Carlos E. R.
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.5 (GNU/Linux)
> Comment: Made with pgp4pine 1.76
>
> iD8DBQFG8qr6tTMYHG2NR9URAqloAJ4+TDS7DztpWeNVOCc1cLyjPXj8UwCghQ1T
> BIuRTiFAj4RcESHHEUFOiiE=
> =zZva
> -----END PGP SIGNATURE-----
>
>   
Objects for the dentry and inode remain in the cache, since there is a 
reference to them. The kernel even keeps objects around for which their 
is no longer a file object. Of course, these entries are discarded if 
there is a need for the memory. You can see all this in /proc/slabinfo.

Bill Anderson
WW7BA
--
To unsubscribe, e-mail: opensuse+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: opensuse+help@xxxxxxxxxxxx