Discover the best of the web!
Learn more about Digg by taking the tour.
Whats new in Linux 2.6.23
kernelnewbies.org — Linux 2.6.23 is about to be released. Here is a readable article explaining whats news. "fairer CFS process scheduler, a simpler read-ahead mechanism, the lguest 'Linux-on-Linux' paravirtualization hypervisor, XEN guest support, KVM smp guest support, variable process argument length, make SLUB the default slab allocator..."
ssam- 597 diggs
- digg it
- anmol2k4, on 10/11/2007, -14/+2Their priority no 1 is Servers. Con is right.
- sunbiz, on 10/11/2007, -1/+7First read this: http://kerneltrap.org/node/14008?page=1
You must have read the link in which Linus puts his point on using the CFS. I think hez got a valid point about the need for stability and I think CFS is more stable. SD may have more promise, but then Linus is practical and doesnt rely on promise alone!! - daftman, on 10/10/2007, -2/+7Con isn't a kernel maintainer. He can promise alot but doesn't deliver.
- eean, on 10/10/2007, -6/+3He maintained his own patchset that was widely used. Thats the closest there is to being a "kernel maintainer" short of Linus himself.
- daftman, on 10/10/2007, -0/+2So? that's mean nothing. He's not a team player. I can maintain my own code too but if it has little relevance to the main tree, it's useless. Don't compare Con with Linus. Different people have different jobs.
- eean, on 10/10/2007, -6/+3He maintained his own patchset that was widely used. Thats the closest there is to being a "kernel maintainer" short of Linus himself.
- shethinkmefunny, on 10/11/2007, -2/+11Dude shut up about Con. He perfectly proved Linus' point about needing a reliable maintainer when he threw a tantrum and quit over his code being rejected.
Con isn't cut out for kernel development. GTF over it.
- sunbiz, on 10/11/2007, -1/+7First read this: http://kerneltrap.org/node/14008?page=1
- 4DFX, on 10/11/2007, -0/+6Can somebody please answer this once and for all:
Does CFS mean better performance?- anmol2k4, on 10/11/2007, -17/+1No, that is why we should have pluggable scheduler.
- sunbiz, on 10/10/2007, -1/+10Yes, CFS does mean a lot more faster performance compared to the current process schedular in Linux. Although, the performance of SD and CFS is the same... I think CFS is more stable now, but may be in the future SD will perform better!!
- daftman, on 10/11/2007, -1/+18No pluggable scheduler is dumb as it reduces performance. If you want different scheduler, recompile your kernel and add it in.
Honestly, how many time do you actually plan to change a scheduler if it is pluggable?- eean, on 10/11/2007, -5/+2Once per computer... some computers are servers, some are desktops. They have very different requirements.
Its ridiculous that sometimes music skips on modern computers, but it does in Linux occasionally (not for me, but there's a few threads on the Gentoo forums about it).- daftman, on 10/10/2007, -0/+4Then have 2 different kernels as there are 2 different schedulers. Deploy them accordingly.
- VenTatsu, on 10/11/2007, -0/+10You confuse pluggable with selectable. Pluggable implies that it is select at load time or at runtime. Selectable implies that it can be selected. Pluggable would cause a slowdown due to requiring indirection in the layer between the kernel and the pluggable module. (Static kernels are slightly faster but bigger than module kernels that load only exactly the same components) A pluggable scheduler is a bad idea as the scheduler must be very very very fast, a selectable at *compile* time scheduler could be useful.
The CFS scheduler actually made many changes to the code that move it towards having selectable schedulers (vs either the old O(1) scheduler or the SD scheduler). Had Con chosen to rewrite the interfaces to the SD scheduler to the new scheduler interface it's possible that it could have been a second selectable scheduler in the kernel at some point in the future, but now that is a moot point, he took his ball and went home because he didn't like the refs call. - sunbiz, on 10/11/2007, -0/+1Well said.. I think Con should have tried to co-operate and make the compatible interfaces... That we ppl who want the SD could have used it!! Now hez got angry like a kid and swollen his face... But I think Con will come back after the community coaxes him a little more!!
- eean, on 10/11/2007, -5/+2Once per computer... some computers are servers, some are desktops. They have very different requirements.
- ssam, on 10/11/2007, -0/+12it depends what you mean by performance. it can't make single tasks run faster.
what it can do is stop one task blocking another. hopefully CFS will mean that interactivity is best. so when you click something it gets processor time quickly, rather than waiting for something else to finish.
there is a lot of disagrement over the best method to improve things. - cronot, on 10/13/2007, -0/+10CFS, as ssam noted, means better interactivity between processes / tasks. That means your system will seem faster when you're having a lot of programs running at the same time, and they are all doing something (as opposed to sitting idle). Some might perceive this as better performance, but technically, I think this is actually better multitasking.
- cbuddha42, on 10/10/2007, -1/+3For the average desktop user CFS pretty much guarantees better "performance". It doesn't speed up your CPU, but it does make it use its time more efficiently. When people say that it's bad or not the best choice the aren't comparing it to the currently scheduler, they are comparing it to another new scheduler by Con. Lots of people think Con's scheduler is better, but CFS is the one that is getting merged. Both beat the pants off the current scheduler.
- anmol2k4, on 10/11/2007, -17/+1No, that is why we should have pluggable scheduler.
- deviantlinux, on 10/11/2007, -4/+5site is down. awesome.
- djGentoo, on 10/11/2007, -2/+6Welcome to Digg.
- djGentoo, on 10/11/2007, -1/+5I tried out CFS back at rc2 with some amazing results. Sauerbraten FPS scored increased by about 20%. And get this: CFS is going into Fedora 8. w00t!
- djGentoo, on 10/10/2007, -2/+1s/scored/scores. doH!
- Sabretou, on 10/11/2007, -0/+11The page is down. Duggmirror doesn't have it. Google, Coral and Wayback don't have it. Digg prevails.
- csshyamsundar, on 10/10/2007, -2/+4Probably the kernel can have both the schedulers and based on distro; one can choose the way they awnt their scheduler.
- jdrivein, on 10/11/2007, -6/+3Great idea: pluggable schedulers!
...- cbuddha42, on 10/11/2007, -0/+4No, pluggable means it is chosen at runtime. There is nothing stopping you from using a different scheduler when you compile.
- jdrivein, on 10/11/2007, -6/+3Great idea: pluggable schedulers!
- bsolidgold, on 10/12/2007, -7/+1http://www.duggmirror.com
- stuy486, on 10/10/2007, -0/+2Did you actually try that before you posted it?
I'm going to go with no since it didn't catch the page.
- stuy486, on 10/10/2007, -0/+2Did you actually try that before you posted it?
- ostracize, on 10/10/2007, -0/+2It's slow, but here's the google cache if it goes down for good:
http://72.14.253.104/search?q=cache:http://kerneln ... - Ansible, on 10/14/2007, -1/+72.6.23 - for hott "linux-on-linux" action!
- lorenzk, on 10/14/2007, -3/+0Now I have a paravirtualization hypervisor.
Ho! Ho! Ho! - baalzebub, on 10/10/2007, -0/+1Enable The Block Layer > IO Schedulers > Default I/O Scheduler > CFQ
Linux gnu 2.6.23 #1 Tue Oct 9 17:18:49 CDT 2007 i686- vicaya, on 10/12/2007, -0/+1Duh, completely different thing. CFQ is an io scheduler that's been around for ages. CFS is the new politically charged thread/process scheduler.
- cawpin, on 10/11/2007, -7/+4I wonder if this will fix the Ubuntu broken CD-ROM problem. Mine still doesn't work properly. "scd0 is not a valid location"
- cawpin, on 10/11/2007, -3/+3WTF? Why am I being dugg down? It's a legitimate question. It was thought to be a kernel level bug.
- cbuddha42, on 10/11/2007, -3/+4No. Changing the scheduler doesn't affect the ability of things to run on your computer. All it changes is the order which processes get CPU time when multiple processes want CPU time.
- ssam, on 10/15/2007, -1/+3thats not to only thing that changed. read the article please
- SteveMax, on 10/10/2007, -0/+1This doesn't look like a kernel problem at all. Could be in anything from scripts to udev, but doesn't look like the kernel.
- cbuddha42, on 10/11/2007, -3/+4No. Changing the scheduler doesn't affect the ability of things to run on your computer. All it changes is the order which processes get CPU time when multiple processes want CPU time.
- cawpin, on 10/11/2007, -3/+3WTF? Why am I being dugg down? It's a legitimate question. It was thought to be a kernel level bug.
- gamelord12, on 10/11/2007, -0/+1Umm...I don't understand any of this, but I assume it's good?
- Darkhacker, on 10/11/2007, -0/+3I don't get the technical details, but I'll try to explain the gist of it.
Completely Fair Scheduler (CFS) - A computer's CPU can only do one thing at a time but it can do it billions of times in a second. The schedulers job is to break up the CPUs resources to the different processes. Even though the CPU can only do a single thing at a time, the reason you can play music, compile software, browse the web, chat on IM, and all those other wonderful things at the same time is because the scheduler is switching back and forth between these tasks. Because the CPU can do billions of calculations in a second, it appears to be seamless to humans and gives the appearance of the computer doing multiple tasks at the same time. The new scheduler is designed to be completely fair in that each process gets an equal share of the CPU cycles where as before the scheduler would to try prioritize tasks and sometimes made mistakes in determining which process to give higher priority to and thus lag or performance loss occurred.
On-Demand Read-Ahead - Mostly just performance improvements, but what this does it is reads a file from the disk and puts it into memory ahead of time so that a program can access that portions of the files content quickly. If you have a massive document file (several thousand pages with graphics and such) then you can't load the entire thing into memory. However if you are on page 7, the read ahead function might load pages 6 and 8 even if they aren't being used at the exact moment in preparation of the user clicking the next page or previous page buttons. This avoids unnecessary disk I/O and makes switching pages faster for the user. That's just one overly simplified example but I think you get the gist of it.
fallocate() - One example is that used by virtual machines. If you have a virtual hard disk as a file or say a program that works with very large files, then you need to be sure that the hard drive won't fill up when trying to save that data. This function pre-allocates an empty file. The file may be 100 MB but it is filled with all zeros to show that it is empty. However you can now write to it and know for sure that you aren't going over any disk quotas or filling up a partition. The function fails if it can't allocate the space (much safer than waiting until saving the actual file with real data) and is mostly reserved for large files like the virtual machine disk example I gave.
Virtualization - Wikipedia can probably explain this better than I can. http://en.wikipedia.org/wiki/Virtualization
Variable Argument Length - Mostly for programmers, but it expands the argument list for a function (like printf) so that you can pass a large number of values (not the size of the value) without the computer freaking out and telling you what a horrible person you are (though it may perform unspeakable acts of perversion with your pet dachshund).
ACPI - Blah blah blah, too much crap to get into it here but its responsible for power management and putting your computer into sleep mode. The main difference in 2.6.23 is that instead of the kernel throwing a pile of drivers at your computer and hoping one of them sticks, it can now load the right drivers using udev.
There are also a lot of new drivers and just general fixes of that sort. I think this covers the basics pretty well.
- Darkhacker, on 10/11/2007, -0/+3I don't get the technical details, but I'll try to explain the gist of it.
- sunbiz, on 10/11/2007, -0/+1Read this: http://www.phoronix.com/scan.php?page=article&item ...
Ok..here's some performance comparison to ppl who want to understand how much performance has improved... The multitasking has improved with CFS, but I think single thread/process performance hasn't changed much!! I'm wondering if Con could release a comparative benchmark for the world to judge on Linus/kernel team's decision!! - murlox, on 10/13/2007, -0/+1I saw a few posts on Gentoo Linux forums regarding 2.6.23 and nvidia drivers, so some of you (YMMV) might want to wait before rushing to install 2.6.23.
© Digg Inc. 2008 —
Content posted by