Tech-admin, can you give me an advice?
I am searching for a kernel with less load to CPU. Some people said that Zen kernel decrease load to CPU...so your Liquorix should also do that. Is that true? I see, that your kernel is very good for perfomance, but i'm primaraly need to have less load to CPU's. I am not sure, is Liquorix kernel good for me or not. Back to top |
|||||
damentz, the zen/liquorix maintainer, will have to answer that. Someone here recently did tests on battery loads with liquorix, and liquorix gave the lowest wattage use, but that is for a laptop, and those optimizations probably don't matter for servers. You are correct, liquorix is mainly built for desktop optimization. But maybe someone has some suggestions for actual cpu load optimization.
What you want is the google linux kernel configuration, heh. Too bad that's not public. Or another large data center user that runs their own linux spin. Back to top |
|||||
:: techAdmin wrote :: damentz, the zen/liquorix maintainer, will have to answer that. Someone here recently did tests on battery loads with liquorix, and liquorix gave the lowest wattage use, but that is for a laptop, and those optimizations probably don't matter for servers. You are correct, liquorix is mainly built for desktop optimization. But maybe someone has some suggestions for actual cpu load optimization.
What you want is the google linux kernel configuration, heh. Too bad that's not public. Or another large data center user that runs their own linux spin. Yes, but it's OK also for servers. I really hope that this kernel can help with cpu load optimization. But the question is - is that possible, that this kernel can help on that. What technologies can do it..? I dont think, that Google are good. It's crappy laggy. Nor like Yandex.com. Back to top |
|||||
well, let's see if damentz has something to add.
I wouldn't judge google servers if accessing from Russia, lag time could just be network. They do such huge data processing that there's a lot more going on behind the scenes, but you can be 100% certain that getting the most efficient energy use out of cpu configuration in kernel is critical for them, we're talking about electricity for 100s of thousands of servers after all, and cooling that heat generation. But google doesn't say anything ever about such things so it won't help. Back to top |
|||||
Mofforg, Liquorix is designed for lower latency workloads on a busy system. You may get some undesired behavior on servers if a single user can cause all your disk transfers to halt as the initiation of an app causes preemption in the IO queue. You can still turn off that feature by sending 0 to /sys/block/sdX/queue/ioqueue/low_latency.
As far as CPU load goes, it's worth a shot. I've heard that BFS can help workloads on servers, but it can also not help. BFS is a very load neutral process scheduler, so it shouldn't cause too many problems if it doesn't suit your work load. You can always raise the rr_interval for BFS to squeeze more throughput out of your server in /proc/sys/kernel. It's normally set to 3 msecs in liquorix (default is 6) so raising this to its default value may make a slight difference if context switching is becoming a problem. Also note that if you are getting strange throughput measurements on the disk, you should adjust the dirty ratio thresholds in /proc/sys/vm/. Right now they're set to values that favor lower latency behavior, but with constant file management, it may not be set to the most efficient values for server loads. Back to top |
|||||
:: damentz wrote :: Mofforg, Liquorix is designed for lower latency workloads on a busy system. You may get some undesired behavior on servers if a single user can cause all your disk transfers to halt as the initiation of an app causes preemption in the IO queue. You can still turn off that feature by sending 0 to /sys/block/sdX/queue/ioqueue/low_latency.
As far as CPU load goes, it's worth a shot. I've heard that BFS can help workloads on servers, but it can also not help. BFS is a very load neutral process scheduler, so it shouldn't cause too many problems if it doesn't suit your work load. You can always raise the rr_interval for BFS to squeeze more throughput out of your server in /proc/sys/kernel. It's normally set to 3 msecs in liquorix (default is 6) so raising this to its default value may make a slight difference if context switching is becoming a problem. Also note that if you are getting strange throughput measurements on the disk, you should adjust the dirty ratio thresholds in /proc/sys/vm/. Right now they're set to values that favor lower latency behavior, but with constant file management, it may not be set to the most efficient values for server loads. For now i fully work from root and no one except me have access to servers, so low_latency should be OK. I am sure BFS should help, but maybe there are any other technologies which should help on this... I think, that i'll stay current value for 'rr_interval'. OK, i'll look at that. Additional question: Is there any changelog of Liquorix kernels? Is there are topic where i can the main changes between zen-kernel and Liquorix? Back to top |
|||||
Liquorix is Zen-Kernel in its own form. Heftig and I work on the Zen Kernel sources and keep all our work public. After we're done, I take a snapshot of the master branch and create the Liquorix patch set. I then throw in my own configuration and distribute it with apt.
In other words, Zen-Kernel is the source / backend. Liquorix is a distribution path with multimedia oriented customizations. If you want to see a Liquorix specific changelog, read the Debian package changelog: Example: :: Code ::
$ dpkg -L linux-image-3.5.0-3.dmz.3-liquorix-amd64 | grep -i changelog /usr/share/doc/linux-image-3.5.0-3.dmz.3-liquorix-amd64/changelog.Debian.gz $ zcat /usr/share/doc/linux-image-3.5.0-3.dmz.3-liquorix-amd64/changelog.Debian.gz | less The Zen-Kernel commit changelog is available at git.zen-kernel.org/zen-stable Back to top |
|||||
:: damentz wrote :: Liquorix is Zen-Kernel in its own form. Heftig and I work on the Zen Kernel sources and keep all our work public. After we're done, I take a snapshot of the master branch and create the Liquorix patch set. I then throw in my own configuration and distribute it with apt.
In other words, Zen-Kernel is the source / backend. Liquorix is a distribution path with multimedia oriented customizations. If you want to see a Liquorix specific changelog, read the Debian package changelog: Example: :: Code ::
$ dpkg -L linux-image-3.5.0-3.dmz.3-liquorix-amd64 | grep -i changelog /usr/share/doc/linux-image-3.5.0-3.dmz.3-liquorix-amd64/changelog.Debian.gz $ zcat /usr/share/doc/linux-image-3.5.0-3.dmz.3-liquorix-amd64/changelog.Debian.gz | less The Zen-Kernel commit changelog is available at git.zen-kernel.org/zen-stable Ok, thx. Back to top |
|||||
Hi. You should be able to run the liqourix kernel on Debian 6. If you add the deb-src line for liqourix, you should be able to pull the source package and then use dpkg-buildpackage to build it with gcc 4.4. You can also change the task scheduler frequency to 300 Hz in the config file before compiling to give you better throughput on your servers. However, with the BFS scheduler, you will be stuck having to use full preemption. BFS is designed to be used only with full preemption and using it with any other preemption model (voluntary preemption, no preemption) could give you unexpected results. You could also just download a kernel from kernel.org, apply any needed patches, and build it. Debian's kernel-package package makes doing this really slick. You should be able to find plenty of tutorials online for doing this.
Back to top |
|||||
All times are GMT - 8 Hours
|