Date: Sun, 11 Jul 1999 15:06:13 +0200 From: Stefan Esser <se@mi.uni-koeln.de> To: Thomas Klein <klein@KryptoKom.DE> Cc: hackers@freebsd.org, Stefan Esser <se@freebsd.org> Subject: Re: question: delay of a context - switch Message-ID: <19990711150613.E388@dialup124.mi.uni-koeln.de> In-Reply-To: <37859C90.322349BD@kryptokom.de>; from Thomas Klein on Fri, Jul 09, 1999 at 08:54:08AM %2B0200 References: <37859C90.322349BD@kryptokom.de>
next in thread | previous in thread | raw e-mail | index | archive | help
On 1999-07-09 08:54 +0200, Thomas Klein <klein@KryptoKom.DE> wrote: > Dose anyone know how long a the kernel is busy with context switching > (beetween two processes) ? > Has anyone tested this yet? > I estimate of about 7 usec duration for that, (on a Pentium 400) but > I think that's to long. There is a port of Larry McVoy's Benchmarks in benchmarks/lmbench, which (as one small part of the benchmark suite) accurately measures context switching overhead in a number of different situations (number of active processes, data segment size). The context switch time depends on a number of parameters, it is not a constant of the CPU and OS. I see 6us on my Celeron 300A when there is no active data and there are only two active processes, 9us with 16 active processes, and 33us with 16 processes that each keep 1MB of memory active. SMP kernels have additional task switch overhead, but I can't offer any data before tomorrow, when I'll run those tests on my systems at work ... Gruß, STefan To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19990711150613.E388>