2.4.19 Kernel To Be Out Soon!

[Armitage]

I finally got around to upgrading my work machine to 2.4.18, so 2.4.19 should be out within a week

Ok, more seriously ... anybody have any experience with the preempt & O(1) scheduling patches?

 

[Electrode]

You troll!

Anyway, to answer your question, I'm running the preemptable kernel patch on all of my rigs. For servery things like apache, php, proftpd, and the like, it's not worth the 12 seconds it takes to apply the patch. However, for a desktop on which you'll be playing games, watching divx movies, and stuff like that, it IS worth the aforementioned 12 seconds. The preempt-kernel patch alone gives me a 7 FPS boost in Quake 3.

Anything in particular you wanted to know?

 

[Armitage]

Originally posted by: Electrode
You troll!

Anything in particular you wanted to know?

Well, I just crashed my box trying to run an app that wanted > 1GB RAM, when I only have 512 MB ram, 512 MB swap. By the time I realized what was happening, it was so unresponsive that I couldn't kill the offending process and had top hit the power button

I'm hoping the preempt patch might help prevent that!

Beyond that ... how's the stability? If it has power, I expect my machines to stay up indefinetely. So today's episode was kind of troubling, but I don't want more trouble by going with this beta kernel patch!

 

[Nothinman]

2.4.19-rc3 was put out recently, so you're probably not far off =)

Well, I just crashed my box trying to run an app that wanted > 1GB RAM, when I only have 512 MB ram, 512 MB swap. By the time I realized what was happening, it was so unresponsive that I couldn't kill the offending process and had top hit the power button

I'm hoping the preempt patch might help prevent that!

Doubt it would help that situation, that situation is the reason Linus recommends swap = memory x 1.5 =)

What may help would be either the -rmap or -aa patches, although all you're going to get in any of the cases is the OOM killer killing something.

 

[Electrode]

No instability caused by the patch. The only time I ever seen Linux crash, patch or no patch, was when I had a bad stick of RAM. Barring such a failute, nothing can make it crash.

Also, nothing is going to keep your system from slowing to a crawl when an app sucks up all of your RAM. It will just slow down and slow down and slow down until the app has no more RAM to expand into, and then the kernel will terminate it and you'll be back to full speed. Trust me, I've had it happen.

The solution is to always have lots of swap space. I have 1 GB of RAM and 3 GB of swap in this rig. The memory hogging app will slow down, but not the rest of the system.

 

[Armitage]

Originally posted by: Nothinman
2.4.19-rc3 was put out recently, so you're probably not far off =)
Doubt it would help that situation, that situation is the reason Linus recommends swap = memory x 1.5 =)

What's wierd is, I always run swap = 2x memory. I must have been asleep at the wheel when I set up this box :/
Just added another GB of swap, so maybe that will help.

What may help would be either the -rmap or -aa patches, although all you're going to get in any of the cases is the OOM killer killing something.

Which would be OK. As electrode suggested, eventually the OOM Killer would have given me back my machine. But I had already waited about half an hour, and needed it back now.

 

[Nighthawk69]

/me begins thinking....

I only setup a 1:1 ratio of physical ram to swap file when I installed RH 7.3.... I guess I need to enlarge that....

I'm told from the guys on #redhat that I should make a swap file instead of partitions.... is this what I should do to correct this?

EDIT: ergeorge, sorry for posting a question in your thread--I'd be happy to take it to a new thread if you'd like.

 

[Armitage]

Originally posted by: Nighthawk69
/me begins thinking....

I only setup a 1:1 ratio of physical ram to swap file when I installed RH 7.3.... I guess I need to enlarge that....

I'm told from the guys on #redhat that I should make a swap file instead of partitions.... is this what I should do to correct this?

Swap files are an easy way to add swap space without reformatting.
It goes something like this:

% dd if=/dev/zero of=swapfile bs=1024 count=524288
% mkswap swapfile
% swapon swapfile

The first command makes a 512MB file.
The second sets the file up as swap space
The last command enables the new swap space.
You can add the last command to a startup script.

Does anybody know if there is any performance hit to using a swap file versus a swap partition?

EDIT: ergeorge, sorry for posting a question in your thread--I'd be happy to take it to a new thread if you'd like.

No problem!

 

[Nothinman]

I'm told from the guys on #redhat that I should make a swap file instead of partitions.... is this what I should do to correct this?

It's probably the quickest since there's no partitioning needed.

Does anybody know if there is any performance hit to using a swap file versus a swap partition?

Yes, there is. Because when it has to page to that file it has to go down through the VFS and filesystem layers, unlike paging to a swap partition.

 

[Nighthawk69]

OK, thanks guys!

I setup a 512MB swapfile so I've got 1GB of swap space now counting the swapfile and swap partition.... I also added the last line of those instructions to my rc.local file, so I guess it should startup with Linux now.

Maybe those instructions should be a FAQ?

Thanks!

 

[Armitage]

Originally posted by: ergeorge
I finally got around to upgrading my work machine to 2.4.18, so 2.4.19 should be out within a week

Told Ya