Quoting Rik van Riel <riel@xxxxxxxxxxx>: > On Wed, 20 Apr 2005, Rik van Riel wrote: > > > I have a feeling there is a LOT of interesting research left to > > do for the swap token mechanism. > > Since I also suspect this work is going to be important for > Linux, I set up a wiki page where we can all take notes. > Otherwise we'll just end up having the same discussion over > and over again, without making progress. > > If you end up playing with the swap token code, please write > down your notes so other people can learn... > > http://wiki.linux-mm.org/wiki/SwapTokenTuning > > -- > "Debugging is twice as hard as writing the code in the first place. > Therefore, if you write the code as cleverly as possible, you are, > by definition, not smart enough to debug it." - Brian W. Kernighan Sorry, I didn't get back to you sooner. I had a busy week. Anyway I figured the best way to get anywhere with this project would just be to get the page faults per second or something and use that to measure the vm pressure. I came up with this small algorithm base on how it seems the cpu load is calculated. I'm not sure I got it quite right in saying it's the swap rate per second but I think the faults per second would be best made available as a property of the runqueue structure so it can be cpu specific. Otherwise if this were tried on a network, it would activate the swap tokening on a computer that might only have low vm pressure. Do you think any of that that's a good idea? Although as far as I know the vm pressure should be spread evenly over the network, so maybe it would be better to just measure it over all cpus or was I right to begin with? I don't think there's anything like that already implemented but I don't know the kernel very well yet so I can't be sure. I was thinking of doing something like this: timeSinceLastFault = currentTime - timeOfLastFault; faultsPerSec = 1 / timeSinceLastFault; av_fps = (av_fps + faultsPerSec) / 2; timeOfLastFault = currentTime; That would go inside the page fault function. I'm going to look into doing it this way as part of the runqueue structure for now. I think I should start by moving the strusture definition into sched.h. I also need to check the math behind that algorithm. Am I on the right lines with the runqueue thing? ---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program. -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
