New Kernel 2.4 Development Branch (-mjc)

Ivo writes: "kerneltrap is reporting: Michael Cohen announced to the lkml his intention to begin a new 2.4 development tree. The first release of his -mjc branch includes a number of performance enhancing patches, including Robert Love's preemptible kernel patch, Rick van Riel's reverse mapping patch and George Anzinger's real time scheduler patch. Michael says of this patch, "I feel that there's need for a rapidly developing '-ac [like]' tree, and so, here we go. Feel free to test it""

Comments

Re:Real world impact?

[blakestah]

So how much gain in performance (or apparent performance) should one expect after applying this combined patch? Are the performance gains only applicable under special circumstances? Are they focused more on desktop apps than server?

I doubt you will see ANY performance enhancements with this patch - in fact, under most circumstances, performance will be worse.

The patch MOSTLY addresses a need to have shorter latency responses under linux. So the real benefit will be seen if you, say, run xmms, browse the web on a java intensive site, and do a make -j10 bzImage at the same time. On most machines this will cause xmms to stutter a little - either an audio skip or the text in the scrolling windows will stop and start. With the patch you can expect perfect xmms performance under broader circumstances.

This has the most significant implications for audio and video under linux - things that require short latencies to perform properly. This is questionably the most needed area of improvement for the linux kernel for desktop use.

However, if you time kernel compiles or run lmbench, you'll probably see slightly (but not hugely) worse results. You can expect that changes to address these issues will be incorporated in mainline kernels eventually, although not necessarily in the form that these patches take. Maybe - it will be interesting to see it sort out.

Re:Ugg, try installing ipnat on freebsd.. same thi

[rtaylor]

ipnat, ipf, and various other tools of that nature have modules installed by default.

Looking forward to the day when kernel modules are a 'load on use' resource. That is, if you try to access /dev/snd it'll load sound drivers until it finds one that works and goes with it. devfs will do some of that, and the rc.conf options also do some of it (in 4.3 or so it does for ipnat).

echo 'ipnat_enable="YES"' >> /etc/rc.conf

The above should load the ipnat kernel module and get you on your way at the next reboot.

NOTE: The above statement depends on ipfilter running, so:

echo 'ipfilter_enable="YES"' >> /etc/rc.conf

may be required as well depending on current configuration.

Re:Great, more fragmentation

[swright]

er, G400s are supported fine in Linux... (from the posts above it appears not in FreeBSD)

WindModems are shite because they do all the signal processing in software - so you take a CPU hit and a transfer rate hit. Much better to have a real modem (for like $5 more..) that does it in hardware, no bandwidth loss and no CPU wastage...

Re:this sounds really cool but

[BRTB]

Slackware has always used an unpatched kernel and managed to keep its high stability level intact... (the Reiser patch in Slack8 being the only exception I can remember, and it was unnecessary unless you were using a Reiser FS)

Of course you may not think Slack's a major distro, but there are those who would beg to differ ;)