On Friday 2009-12-04 13:11, Patrick McHardy wrote: >Jan Engelhardt wrote: >> On Thursday 2009-10-29 16:59, Patrick McHardy wrote: >>>> On Wednesday 2009-10-28 16:57, Patrick McHardy wrote: >>>>> Jan Engelhardt wrote: >>>>>> I incorporated your suggestions and propose this new 4-series set. >>>>> Could you give me a short description of the changes relative >>>>> to the first incarnation? Thanks. >>>> http://osdir.com/ml/general/2009-08/msg10540.html >>>> - splitting xt_repldata_create >>>> - giving it a name of your choice >>>> - return semantics (found a missing return keyword; just refreshed top >>>> is commit v2.6.32-rc3-4-g7891b6c) >>>> >>> OK thanks, that leaves my worries about adding additional runtime >>> overhead. I'll do some quick benchmarks to see if it really has >>> a measurable impact. >>> >> How would one measure this in a meaningful way? rdtsc? (that would only >> work on x86, in fact.) >> >> And I guess that one could >> always construct a case whereby the patched result looks 'slow'. :/ > >Sure .. but I don't think it will matter much in comparision to large >rulesets. I usually use get_cycles() in nf_hook_slow() to calculate >the cycles spent in the netfilter hook functions. In this case just >measuring the difference before and after with all affected tables >loaded, but without any rules, seems fine to me. Seems to be promising, > summary(base) Min. 1st Qu. Median Mean 3rd Qu. Max. 15520 15520 15520 15520 15520 15520 > summary(old) Min. 1st Qu. Median Mean 3rd Qu. Max. 11890 17900 21370 20440 22680 53210 > summary(new) Min. 1st Qu. Median Mean 3rd Qu. Max. 11430 14770 17660 18300 20830 34510 What surprises me a bit is that there is no fluctuation whatsoever in the base case. -- To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
