I've noticed that UTM9 just isn't up to par with what i'm going to call UTM8. I've also noticed a ton of swapping being used even with vm.swappiness=0 set. even with 3 gigs of ram installed I was still swapping 25% even though the system was caching more than 25% of physical ram. I've seen this before in older kernels. Back in 2.4 I filed a bug with centos and by extension redhat about a similar problem...it was later fixed in the next kernel patch.
I decided to try something..something i did back in the 2.4x days....I disabled swap entirely. I now have the latest beta running with 3 gigs of ram and zero swap(swapoff -a). I've had this running for 24 hours:
free
total used free shared buffers cached
Mem: 3087332 2730288 357044 0 94552 769160
-/+ buffers/cache: 1866576 1220756
Swap: 0 0 0
Within 24 hours i would have 10% swap...within 36 hours i would have 25% swap. There's no reason for a system that is under ZERO memory pressure to be swapping to disk at a rate of 25% or higher when more than 30% of physical ram is unused. I could go digging into the kernel deeper to find out why this is occurring but this is a development problem..not a partner/user problem. V8 does NOT display this behavior. I've noticed this since the first beta of UTM9. I'm curious how many other performance regressions I've found might be addressed? The testing begins..:)
I decided to try something..something i did back in the 2.4x days....I disabled swap entirely. I now have the latest beta running with 3 gigs of ram and zero swap(swapoff -a). I've had this running for 24 hours:
free
total used free shared buffers cached
Mem: 3087332 2730288 357044 0 94552 769160
-/+ buffers/cache: 1866576 1220756
Swap: 0 0 0
Within 24 hours i would have 10% swap...within 36 hours i would have 25% swap. There's no reason for a system that is under ZERO memory pressure to be swapping to disk at a rate of 25% or higher when more than 30% of physical ram is unused. I could go digging into the kernel deeper to find out why this is occurring but this is a development problem..not a partner/user problem. V8 does NOT display this behavior. I've noticed this since the first beta of UTM9. I'm curious how many other performance regressions I've found might be addressed? The testing begins..:)