On 15/06/2014 6:37 AM, Catalin Salgau wrote: > On 15/06/2014 3:59 AM, Ed Cashin wrote: >> Regarding the idea of reordering the tests in BPF to minimize the number >> of tests---can you measure a performance difference? If so, what do the >> tests and results look like? >> >> Thanks. >> > Largely negligible with usual hardware. > Since the bytecode is emulated or jitted to native code for execution, I > would reckon one would notice a difference in low-powered devices and on > 10G links, where fast-fail would save precious cycles. > (and while we're on the subject of precious cycles, there is basically > duplicate code in aoe.c:aoe(void) after the call to getpkt(). except for > the size check and mac check, the others are found in the BPF bytecode) Addressing testing and results on this - I believe it's very difficult (if even possible) to time. The idea in this change proposal is to throw out useless packets faster. Yielding from filtering faster would benefit the system, but it wouldn't be noticeable in one instance of vblade running on a test system. The advantage should be noticeable under heavy load with a large number instances of vblade running on the same interface, in which case it should free up some CPU from the filter evaluation. > ------------------------------------------------------------------------------ > HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions > Find What Matters Most in Your Big Data with HPCC Systems > Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. > Leverages Graph Analysis for Fast Processing & Easy Data Exploration > http://p.sf.net/sfu/hpccsystems > _______________________________________________ > Aoetools-discuss mailing list > Aoetools-discuss@xxxxxxxxxxxxxxxxxxxxx > https://lists.sourceforge.net/lists/listinfo/aoetools-discuss > ------------------------------------------------------------------------------ HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions Find What Matters Most in Your Big Data with HPCC Systems Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. Leverages Graph Analysis for Fast Processing & Easy Data Exploration http://p.sf.net/sfu/hpccsystems _______________________________________________ Aoetools-discuss mailing list Aoetools-discuss@xxxxxxxxxxxxxxxxxxxxx https://lists.sourceforge.net/lists/listinfo/aoetools-discuss
