On 9/30/21 2:49 AM, Praveen Ponakanti wrote: > I am trying to use the squid cache as an outbound HTTP/S proxy service > for our production network and would like to scale up the request per > sec by running it on multi-core systems. > I have searched up the archives and config docs, but couldn't find > anything more recent than the below on setting up SMP workers. > https://wiki.squid-cache.org/ConfigExamples/SmpCarpCluster > https://wiki.squid-cache.org/ConfigExamples/MultiCpuSystem The two old exotic configurations above are not what you should be starting with when thinking about SMP support in modern Squids. The best starting points I know about are * https://wiki.squid-cache.org/Features/SmpScale * http://www.squid-cache.org/Doc/config/workers/ * http://www.squid-cache.org/Doc/config/cpu_affinity_map/ * http(s)_port worker-queues option That Feature wiki page does not get many updates these days; if there is a conflict between information sources, the directive documentation in your squid.conf.documented may have more recent information. > * Can someone please let me know if the config examples in those docs > still apply to the latest version 5.1. I bet those use cases themselves do not apply well to your situation. > * Do we still need the combination of squid, frontend, backend.conf > files, or is it enabled simply with the "workers" line in the main > config file? The workers directive does not require multiple Squid instances (with multiple configuration files, etc.). > * When I attempt to run the squid server with "workers 4" in the > squid.conf, I get the following error (one for each kid) and the main > process does not bind to the "http_port 3128". I see several UDP ports > opened up for each kid process. > 2021/09/29 00:44:10 kid5| commBind Cannot bind socket FD 11 to [::]: (2) No such file or directory Unfortunately, I cannot tell exactly what went wrong based on that low-level message alone, but it could be a variant of [1]. If you cannot figure it out after checking [1] suggestions, consider sharing "squid -X" startup cache.log for analysis. There is [1] https://wiki.squid-cache.org/Features/SmpScale#Cannot_bind_socket_FD_NN_to_.5B::.5D:_.2813.29_Permission_denied The UDP ports you are seeing are probably for the internal DNS resolver which is not SMP-aware. > * Does anyone have recommendations on the maximum number of workers to > use on a 64 core host (assuming no other CPU intensive apps are running > on the same host). Caching is not a must for our initial deployment, so > we are fine disabling caching. 28-30. See the following wiki section for the corresponding rules of thumb: https://wiki.squid-cache.org/Features/SmpScale#How_to_configure_SMP_Squid_for_top_performance.3F HTH, Alex. _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
