Wed, 15 Oct 2008
Kernel Summit, Day 2
Just a short report from the rest of the Kernel Summit:
Probably the most interesting discussion the second day was about
thread pools. I believe there is no clear solution, because it could
for example make sense to have N
kcryptd threads for N CPU cores
if the most of your workload is in handling encrypted block devices,
but it probably does not make sense to have 2048 kernel threads to handle
networking bottom half on a 2048-CPU SGI box where most of the cores are
too far (NUMA-wise) from the network card, and the box mainly does other
things than handling networking. Maybe some kind of dynamic allocation?
However, moving even IRQ handlers to separate threads could help with
some latency problems like the one I observed on my router.
We have got a development board from Gumstix as a present. It is an interesting toy - a very tiny board with TI OMAP CPU (like the one from Nokia Maemo), memory, and a microSD slot. And a bit bigger daughterboard with I/O ports (several USBs, power, HDMI, sound, WiFi/BT, etc.). I just need to get a proper power brick for the local wall outlets (the one we got was for US outlets).
The joint dinner with the Linux Plumbers Conference: playing Wii Sports with Bartlomiej Zolnierkiewicz, Herbert Xu and Evgenij Polyakov (Wii is cool and definitely very different from other game consoles, try it!), and then discussing dm-crypt threads and Brno Red Hat branch with Alasdair Kergon.
3 replies for this story:
I'm very curious about the conclusions (if any) on Brno RH...
No special conclusion, just discussing the people we both know.
Picture shows back of my head, Jonathan Corbet, Al Viro and another person I didn't know chatting. Ted Tyso in the blue shirt right next to the blue wall.