Comments on: All About Linux Swap Part 3: Analysis http://www.ideaexcursion.com/2009/02/06/all-about-linux-swap-part-3-analysis/ Technology Musings Tue, 09 Mar 2010 13:15:28 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: Taylor Gerring http://www.ideaexcursion.com/2009/02/06/all-about-linux-swap-part-3-analysis/comment-page-1/#comment-151 Taylor Gerring Tue, 10 Mar 2009 21:43:48 +0000 http://www.ideaexcursion.com/?p=424#comment-151 I've been very happy turning swappiness down from 60 to 40. I don't often fill all 4GB of RAM, so this is enough of a change to prevent swap being used unnecessarily without turning it off completely. I’ve been very happy turning swappiness down from 60 to 40. I don’t often fill all 4GB of RAM, so this is enough of a change to prevent swap being used unnecessarily without turning it off completely.

]]> By: Trenton D. Adams http://www.ideaexcursion.com/2009/02/06/all-about-linux-swap-part-3-analysis/comment-page-1/#comment-150 Trenton D. Adams Tue, 10 Mar 2009 18:45:38 +0000 http://www.ideaexcursion.com/?p=424#comment-150 I have used no swap for a couple of years. I was sick of my system swapping so much that I just quit using a swap file all together. At the time, I didn't know that the kernel was configurable in this regard, so I just worked without swap. I have 4G of memory, and I NEVER get out of memory errors. I'm a Java developer, so that's pretty amazing!!! HEHE. Recently I went searching for the configurability of swappiness. It is documented in the kernel source. So, I'm trying out using swap again, but with a swappiness of 0. I haven't seen any problems yet, and I honestly don't expect any. So yeah, I think Andrew Morton's view is a legacy one. I understand his point, and I would have agreed with 256M of RAM, but things really have changed. 4G is common place these days. I think with swap disabled, or swappiness set to 0, the kernel does an extremely good job of managing system cache. The system cache is always high UNLESS the kernel needs to allocate more memory. I'm usually seeing >= 1G system cache, with a few M free, and I appreciate that. As for the kernel killing processes, when out of memory, I don't have a problem with that at all. I just think the default behaviour for that is, well, REALLY BAD. The default is to kill the one with the most memory. Well, on a server, that's a REALLY BAD THING. Suddenly a service could be down, and you didn't know. So, I like to change the default behaviour like so... echo 1 > /proc/sys/vm/oom_kill_allocating_task This will cause the allocating task that ran out of memory to be killed. Usually, this is what I want, *I think*. However, I am uncertain of the behaviour if you are really at your memory peak. But, I'm trying it out, and I'll see what happens. I have used no swap for a couple of years. I was sick of my system swapping so much that I just quit using a swap file all together. At the time, I didn’t know that the kernel was configurable in this regard, so I just worked without swap. I have 4G of memory, and I NEVER get out of memory errors. I’m a Java developer, so that’s pretty amazing!!! HEHE.

Recently I went searching for the configurability of swappiness. It is documented in the kernel source. So, I’m trying out using swap again, but with a swappiness of 0. I haven’t seen any problems yet, and I honestly don’t expect any. So yeah, I think Andrew Morton’s view is a legacy one. I understand his point, and I would have agreed with 256M of RAM, but things really have changed. 4G is common place these days.

I think with swap disabled, or swappiness set to 0, the kernel does an extremely good job of managing system cache. The system cache is always high UNLESS the kernel needs to allocate more memory. I’m usually seeing >= 1G system cache, with a few M free, and I appreciate that.

As for the kernel killing processes, when out of memory, I don’t have a problem with that at all. I just think the default behaviour for that is, well, REALLY BAD. The default is to kill the one with the most memory. Well, on a server, that’s a REALLY BAD THING. Suddenly a service could be down, and you didn’t know. So, I like to change the default behaviour like so…

echo 1 > /proc/sys/vm/oom_kill_allocating_task

This will cause the allocating task that ran out of memory to be killed. Usually, this is what I want, *I think*. However, I am uncertain of the behaviour if you are really at your memory peak. But, I’m trying it out, and I’ll see what happens.

]]> By: stan_qaz http://www.ideaexcursion.com/2009/02/06/all-about-linux-swap-part-3-analysis/comment-page-1/#comment-93 stan_qaz Sat, 07 Feb 2009 00:08:55 +0000 http://www.ideaexcursion.com/?p=424#comment-93 Don't overlook the need for swap if using some of the hibernate options. This was briefly mentioned in part 1. Don’t overlook the need for swap if using some of the hibernate options. This was briefly mentioned in part 1.

]]>