Arthur Corliss wrote:
> On Sun, 27 Sep 2009, Shane Spencer wrote:
>
>> Most modern distro configs specify the HZ as well as the NO_HZ config
>> variable. Chances are changing from 1000 to 300 will do nothing since
>> NO_HZ talkes precedence AFAIK.
>>
>> There's lots of good info on the webs on CONFIG_CC_OPTIMIZE_FOR_SIZE
>> and the benefits it offers many targets. The kernel I'm using doesn't
>> enable it because it's expected to run on systems with P4 like chips
>> with a large cache. I don't know how large the executable is once
>> loaded, but the compressed kernel image is only 3.4M.
>
> HZ still has an effect, even in tickless kernels. Desktop systems and
> virtualization hosts should tend towards high HZ values in order to reduce
> latencies. General purpose servers and VM hosts, however, should tend
> towards lower VM values.
I had no idea - http://kerneltrap.org/node/6750 explains it in more
detail it appears. Thanks Arty!
> I always use size optimization myself. Cache coherency in general is a big
> deal for performance, even more so when you're running multiple VMs. And
> most people don't understand that gcc's "performance" optimization can not
> only introduce some bugs, but kill you with much larger kernels (especially
> optimizations like loop unrolling), which, again, devestates cache
> coherency. Which is why many projects rarely use higher than '-O2'.
>
> --Arthur Corliss
> Live Free or Die
-- Attached file included as plaintext by Ecartis --
-- File: signature.asc
-- Desc: OpenPGP digital signature
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iEYEARECAAYFAkq/plQACgkQXK/vGhypreJ/NACfZTJ+XpJwxVlYVp2ZfW9DoN5k
+qUAnRoidrmPUsAyB3dHrHyrlxdubJ+3
=ldor
-----END PGP SIGNATURE-----
---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
Received on Sun Sep 27 09:53:18 2009
This archive was generated by hypermail 2.1.8 : Sun Sep 27 2009 - 09:53:18 AKDT