Siduction Forum
Siduction Forum => Scripting & Kernelhacking => Topic started by: ralul on 2012/03/31, 14:38:17
-
Finally after 175 patches there is linux-3.3.1-rc1 availlable on kernel.org. I have testet successfully linux-3.3+172patches for now two days.
But some days before I really had a breakdown when running the 3.3-kernel with the 132-patches in the stable-queue:
There was a very high sound and then hardly accessing harddrives. I booted back to linux-3.2 to test my hardware if it was broke, but all ok. I should have looked into logs but I didn't, so I dont know nothing about what made this ...
But now it looks fine to me
without BFS-Scheduler patch: No one of us had missed BFS?
This BFS patch wasn't ready to apply for towos first linux-3.3 release. There is a debate going on if this scheduler just feels good - like a self full filling prophecy ....
-
There is a debate going on if this scheduler just feels good - like a self full filling prophecy ....
I have always wondered why the BFS is such a topic of hot debate -- is it not possible to make some benchmark tests and give some real data, for users to evaluate? It seems a little like audiophiles arguing about monster cables and what they hear.
Personally, I have been very happy with the deadline scheduler for the past year -- it works very nicely on my SSD-based systems.
-
@dibl,
deadline scheduler is an alternative to BFQ or CFQ, like
CONFIG_IOSCHED_CFQ=y
this is about io queuing.
Con Kolivas made the BFScheduler which is about tasks distributed amoung cpu-cores. Latencies of task schedulers are hardly measurable. You really can test performance throughput of tasks easy but latency is not that neutral to measure:
A human beeing can not feel the difference of waiting 32 or 31 milliseconds. But if you have to wait once more than a second you will be disturbed. This is the "worst case" you can feel . This case can be hidden in statistics.
-
Thanks ralul.
I think I will stay happy with my current setup. CPU running at 4.2GHz will keep a smile on my face, regardless of the scheduler. :)
-
@dibl, this my post sceptical against BFS as a scheduler was just a provocative act to get some reactions of our community. But it turns out nobody interested :(
In fact, after getting patch number 176 in the stable-queue, I recompiled with BFS enabled, I am convinced
everybody could feel the difference!
As I said, out there in the mailing lists there are discussions if BFS is better with latencies are just feelings of biased people. I would favor some "double blind" experiments in our community, but it seems nobody really interested ...
-
Well, actually I would test a kernel with BFS if one would be avaible. But on the downside I don't have a siduction installation on real hardware(only in a vm... yes, shame on me :D ) and a test in a vm woudn't make any sense to me...
-
@MadDin,
if you had installed linux-image-3.2-XY.towo.X-siduction-ARCH you have tested BFS.
Today linux-3.3.1 released after two weeks of heavy bug fighting.
Interesting news about coming linux-3.4-rc1:
x32-abi will be available for amd64 architectures!
This will reduce ram requirements for applications using a lot of pointers. I think this also effects C++ programs which have a lot of heritage to point to, like libreoffice?
As ram usage normally goes up this will be a rare feature against a longtime trend. I guess about 10 to 30 percent less ram usage for applications affected?
-
@MadDin,
if you had installed linux-image-3.2-XY.towo.X-siduction-ARCH you have tested BFS.
Today linux-3.3.1 released after two weeks of heavy bug fighting.
Interesting news about coming linux-3.4-rc1:
x32-abi will be available for amd64 architectures!
This will reduce ram requirements for applications using a lot of pointers. I think this also effects C++ programs which have a lot of heritage to point to, like libreoffice?
As ram usage normally goes up this will be a rare feature against a longtime trend. I guess about 10 to 30 percent less ram usage for applications affected?
Why not build experimental kernels for those who want to try this cutting edge stuff? That x32 abi stuff sounds good for those who need 32 bit apps under a 64 bit kernel and having them run with better performance and less memory usage
-
@DeepDayze, you already can run 32-bit apps, like google-earth:
http://forum.siduction.org/index.php?msg=18619#18619
x32 is a amd64-featured sub-architecture using half sized pointers, which means in this case max 3Gigabyte address space per application.
-
@DeepDayze, you already can run 32-bit apps, like google-earth:
http://forum.siduction.org/index.php?msg=18619#18619
x32 is a amd64-featured sub-architecture using half sized pointers, which means in this case max 3Gigabyte address space per application.
yes I know its possible to run 32 bit apps already on x86_64, and will x32-abi work out of the box with 32 bit apps or will these apps need recompiling?
-
x32-abi compiled apps are real amd64 compiled apps but with 32bit pointer lenght. This is totally different from known arch-x86 32bit apps...
-
Well I will test BFS on my Gentoo:). I think Zen-Sources have BFS already included or do you know an alternativ?
-
MadDin, alternative:
apt-get install quilt
cd /usr/src/linux
mkdir patches
cd patches
wget \
http://ck.kolivas.org/patches/3.0/3.3/3.3-ck1/patches/3.3-sched-bfs-420.patch
echo "3.3-sched-bfs-420.patch" >series
cd /usr/src/linux
quilt push -a Of cause, you can elaborate with more patches this scheme ...