> > Maybe the concept of swapping is becoming obsolete? > > I think its definitely headed that way, if not already there for most > new systems take a deep breath. this is manifestly untrue. in the current era, swap is a way to optimize the use of memory. please don't even think of its original meaning as a place to write out a process's complete VM in order to switch to another process's VM. swap is where the kernel puts less-used pages. having lots of memory does NOT magically mean that all pages are equally well-used. if you do actually have so much physical memory that the kernel never has any shortage of memory, well, bully for you, you've wasted big bucks. and it is waste. consider the page of my Firefox browser that contains code for interpreting devanagari unicode glyphs. it doesn't get much use, and so it should get swapped out. writing out a swap page is basically free, and it means I have one more page available for something more useful. since my frequency of reading devanagari pages is sort of low, that page can be used to cache, say, http://images.slashdot.org/title.gif. swapping is how the kernel optimizes space used by idle anonymous pages. if you understand that, it says absolutely everything you need to know: that swapping will always have some value, that it depends on high variance in the "temperature" of pages (and thus the concept of kernel memory "pressure"), that it doesn't apply to a system with no dirty anonymous pages. you can refuse to do this optimization, and your system will run poorer. the kernel can do this optimization poorly, and thus slow down, not speed up. you can waste your money on so much ram that the kernel never experiences memory pressure (but you'd be surprised how much ram that would require!) swapping is nearly free, since doing an async write doesn't slow anything else down (assuming good page choice, otherwise lightly loaded disks, nothing misconfigured like PIO.) swapping is a recognition that disk is drastically cheaper than ram. swapping does mean that idle anonymous pages could be corrupted by a flakey storage subsystem - but if that happens, you have other serious issues - what does it say about the pages comprising the text of your applications? remember that a big, hard-pressed system might have, say, a gigabyte of swap in use, but even a small desktop will have 50x that much in files exposed to the same hypothetical corruption. yes, I configure my systems with un-raided swap, because if disks are *that* flakey, I want to know about it. finally, ram prices (per GB) have been mostly stable for a couple years now. it's true that ram is faster, and systems are slowly increasing in ram size, but nothing dramatic is on the horizon. I run a supercomputer center that is buying a large amount of new hardware now. the old systems are almost 4 years old and have 1GB/cpu (4-way alphas). new systems will average about 4GB/cpu, and only a few of our users (HPC stuff way beyond desktops) would like more than 8GB/cpu. all of the ~6K cpus we buy will be in systems that have swap. regards, mark hahn. - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
