Cost of user-kernel mode switch
 

Hi,

I am working on a high performance proxy service and in the process was trying to find out the cost incurred in "user -> kernel" mode or "kernel -> user" mode transitions. If the cost is not significant, I could the existing kernel stack else will have to work on a user mode driver.

Would appreciate if anyone has some insights, benchmarks for the same or can sugget a good way to benchmark this.

Thanks,
Dev

Hello,

Well I am not an expert in this but it seems that any literature you will find on this topic will state that user/kernel mode switch is costly. But it seems difficult to find benchmarks.
Anyway I answer you just to share with you this link: http://web.yl.is.s.u-tokyo.ac.jp/~tosh/kml/
I hope you'll find it useful.

Regards,

Oliv'

Thanks Oliv. There is some latency associated with mode switch, though the consensus is that its not as expensive as it used to be.
I was trying to get an estimate on what the cost is, relative to actual processing.
I did some tests myself with a 2Ghz processor, a single mode switch seems to be taking 150 cycles, so looks like not that bad.

Hello,

yes that's not so much time... also it will depend on how many times your application will do some context switch ;)
By the way, as I am curious, I have continued my search on the subject and found this page: http://blog.tsunanet.net/2010/11/how...e-context.html where you will find interesting benchmarks about context switch.

Regards,

Oliv'