sony-vaio-z-series team mailing list archive

["Jon Young" <jon@xxxxxxxxxxxxxxxxxxxx>]

> -----Original Message-----
> From: Michael Zugelder [mailto:michaelzugelder@xxxxxxxxxxxxxx]
> Sent: Saturday, January 08, 2011 11:39 AM
> To: Jon Young
> Cc: sony-vaio-z-series@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Sony-vaio-z-series] getting kubuntu 10.10 installed
> 
> Hi,
> interesting message, I have a few remarks/questions.
> 
> On Sat, 2011-01-08 at 11:03 -0500, Jon Young wrote:
> > run guided partition creation with LVM encryption passphrase
> 
> This is interesting, since earlier 2010 (Ubuntu Lucid with kernel
> 2.6.32) I hit an extremely annoying kernel bug, that in kernels >=
> 2.6.34 kills my SSD performance when using dm-crypt.
> I filed a bug report, but it seems the combination of SSD and
> encryption
> is very rare (the impact is much less dramatic with an HDD).
> Link: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/653611
> I would appreciate it if you could confirm/deny the existence of this
> problem. A quick test with palimpsest should do the trick.
Since I set this up straight out of the box I didn't realize the performance hit but my test actually shows this is considerably worse than your numbers but I don't have old kernels or unencrypted to compare against.  I've attached the graph (not sure if the attachment will be stripped for the list though) against dm-4 which is my mounted luks encrypted root partition.  Note that this model is dual 64G SSDs.  Out of curiosity I also bench'd my /boot which is of course unencrypted.  The graph I think ended up a bit skewed because of the small partition size and some caching but just looking at the minimums is an enormous difference.  I had no idea my performance was such a small fraction of what it could be.


> 
> Please note that the severity of this latency issue correlates with the
> CONFIG_HZ kernel config. You can use `grep CONFIG_HZ= /boot/config-*`
> to
> get the value for each installed kernel. 100HZ (Ubuntu default) results
> to ~10ms latency, 300HZ and 1000HZ give me ~3ms, 100HZ with an old
> kernel (2.6.32) achieves ~0.2ms.
My kernel is set to the 100 default.

<snipped>

> > 4-5hrs of battery on speed with wireless off and browser closed
> (flash is a battery killer) and no additional tuning
> 
> 4-5 hours with NVidia graphics and the resulting maximum brightness?
> I barely get down to 12W with minimum brightness, only Intel graphics,
> no wireless and idling, which would result in a 4h55m battery lifetime.
Yes, but I'm extrapolating from what powertop is telling me and regular usage during which I haven't gone below 40% charge - I didn't run a proper battery rundown test so my data may be way off.

You now have me very interested in the bug you've found.  I have a Dell E6410 with a 256G SSD in the office that is not RAID of any sort or encrypted that I'll test (as is) on Monday for curiosity.

Thanks,
Jon

Attachment: boot benchmark.png
Description: boot benchmark.png

Attachment: root benchmark.png
Description: root benchmark.png