On 04/26/2011 03:03 PM, Manish Kumar wrote:
It's no hard and fast rule that swap space should be 2xRAM size.
Thats partially true. The idea of 2x<ram> came from the linux-2.2 kernel VM and how things were handled in those days ( and I suspect before then ); Having 2x<ram> actually made some of the swap ops faster in/out and there should be plenty of quantifiable evidence for that around.
thinks have changed quite a lot since those days. with a newish 2.6 kernel >= 8 GB of ram and the sort of app one might be running its worth considering what you want to achieve with the swap. In most cases, unless you are running nearline ssd at 400MB/sec or better +rw - swapping into 4+ GB is usually 'bad' for the overall machine state. Some swap usage will happen, and unless you are trying to fix a real issue - just going with system distro defaults is considered reasonable.
btw, most sysadmins these days keep swap limited to 4 or in some cases 8 gb, but still consider aligning it with real physical ram. So 10GB on a 48GB ram machine sounds off. I'd recommend 8 or 12 instead.
w.r.t tmpfs, its allocated and managed dynamically. So its not usually a big concern. However, why use tmpfs at all !
- KB _______________________________________________ Ilugd mailing list Ilugd@lists.linux-delhi.org http://frodo.hserus.net/mailman/listinfo/ilugd
