On 11/02/2013 10:13 PM, Dr.x wrote: > Alex Rousskov wrote >> http://wiki.squid-cache.org/Features/SmpScale#Will_similar_workers_receive_similar_amount_of_work.3F >> >> (but that is not the only theory in existence and YMMV). >> >> HTH, >> >> Alex. > > > hi alex , thanks alot for reply , > > actually it was miss config , i was supposed to put visible hostname for > other processes of squid ! > > again , > relative to core/worker distribution , > > actually i visited > http://wiki.squid-cache.org/Features/SmpScale#Will_similar_workers_receive_similar_amount_of_work.3F > more and more , abut not understanding it 100 % > > > i didnt find clear answer of " will similar workers have similar cpu ?? " , The answer is "Whether similar workers are going to receive the same amount of work depends on your OS environment". > but as i understood , that developers made a patch tat optimize the cpu > distribution , True. > but also with result you put , there were clear difference of workers cores > utiliztion The posted results from a patched Squid show an improved core balance: The utilization of the four busiest cores with unpatched Squid (first table, first column) differed by 60% [(20-8)/20]. Patched Squid (second table, second column) shows 19% difference (). Is 19% difference perfect? No. A better patch or a kernel fix would be needed to improve the balance further IMO. > does from you result , mean that we cant have full load core distribution ? Due to regular traffic variations (see Amos' response for details), the difference is unlikely to become zero, if that is what you mean by "full load core distribution". Please also note that the first column in the second table (patched Squid) shows instantaneous CPU core utilization and is not expected to show similar numbers for workers. Hope this clarifies, Alex.
