On 2016-03-31 18:44, Alex Rousskov wrote: > > My working theory is that the longer you let your Squid run, the bigger > objects it might store in RAM, increasing the severity of the linear > search delays mentioned below. A similar pattern may also be caused by > larger objects becoming more popular during certain days of the week. > Thanks. The current maximum_object_size_in_memory is 19 MB. > > In summary, dealing with in-RAM objects significantly larger than 1MB > bigger the object, the longer Squid takes to scan its nodes. > > Short term, try limiting the size of in-RAM objects using > maximum_object_size_in_memory first. If that solves the problem, then, > most likely, only cached objects are affected. I'll try this next, after I've given Yuri's suggestion a while to yield results (or not). > Also, forcing shared memory cache (even if you are not using SMP Squid) > shared memory code itself does not have the above linear search, SMP > Squid still uses local memory code and might hit the same linear search. I'll look at that too. > guess less. Please point to or copy to this email exchange to your bug > report. I've added bug http://bugs.squid-cache.org/show_bug.cgi?id=4477, and included a reference to this thread at the end of the bug report. Thanks for the assistance. Mark _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
