Messages in this thread | | | From | Matt Heler <> | Subject | Re: excessive swapping | Date | Fri, 13 Aug 2004 02:00:02 -0700 |
| |
Can you try using Nick Piggin's np patchset ?
Matt H.
On Friday 13 August 2004 12:10 am, Florin Andrei wrote: > On Thu, 2004-08-12 at 23:44, Florin Andrei wrote: > > On Thu, 2004-08-12 at 23:40, Florin Andrei wrote: > > > The system is swapping excessively. There's no way the total size of > > > the applications exceeds the size of RAM. There's plenty of room to > > > spare, yet 16% of the 530MB of swap is used. > > > > Now it's 22% and counting. Way to go. :-( > > Now it's 27%. You get the picture. > > Anyway, out of the 512MB of RAM, like 390MB are disk cache. No wonder > that the useful pages are swaped out. > > It seems like the kernel believes that the disk cache has some > miraculous properties w.r.t. the system performance, and desperately > tries to grow it as much as possible. > This is wrong religion. The reality is opposite. The system is much > slower, because applications are thrown out in the swap, then sucked > back in, which is a very slow process. > > The efficiency of increasing the disk cache decreases exponentially with > size, like any other cache. Then what's the point of sacrificing useful > memory just to increase some hypothetical "useful" cache? > > Even on a server, the same universal laws still apply, the efficiency of > increasing cache still decreases exponentially. There's still precious > time wasted when an application is sucked back in from swap, at the > price of an immeasurably small performance gain due to the disk cache > being larger. > > I'm sorry for rambling, but to me the current swapping policy is so > blatantly wrong. Besides the space occupied by the apps themselves, > there's a lot of room to _spare_ - then why swap? - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |