On 10/10/2013 09:32 AM, Alex Rousskov wrote: > On 10/10/2013 06:11 AM, Amos Jeffries wrote: >> On 11/10/2013 12:13 a.m., Alfredo Rezinovsky wrote: >>> El 10/10/13 05:07, Ding Guigeng escribió: >>>> now my squid server running with 3 workers well.but the usage of per >>>> core is uneven. > >>> The load balancing in the workers case is done by the kernel >>> And yes, is't more distributed than "balanced" > >> It is load balanced by connection (TCP SYN packet). > > > It is not in many kernels, unfortunately. This is discussed in detail at > > http://wiki.squid-cache.org/Features/SmpScale#Will_similar_workers_receive_similar_amount_of_work.3F Ignore my comment. Amos' comment is accurate, and the SmpScale imbalance does not apply to this case. I was confused by the "workers" terminology used in this thread instead of "instances". To me, "workers" imply SMP, but the "Frontend Balancer Alternative 1: iptables" setup discussed in this thread uses essentially independent Squid instances rather than identically configured SMP workers. Those independent instances can be implemented using an SMP configuration with a workers directive, but I do not consider such an implementation a supported use case (some do). Sorry for the noise, Alex.
