On 10/01/19 1:54 am, Ralf Hildebrandt wrote: >> It is most useful for peer connections where the peer selection >> algorithm directly depends on the RTT times. > > Like reverse proxies? > Like weighted round-robin, ICP, HTCP. >> It is also used with persistent connection selection as a small >> optimization to prioritize faster connections to be selected for re-use >> rather than slower connections. This makes a big difference for servers >> operating behind an IP load balancer. >> >> This feature is always an optional one. In absence of NetDB RTT data >> algorithms using it assume that all TCP connections are equally fast. > > In a "normal" proxy setup (users are behind 4 proxies which connect > them to the Internet), does use of Netdb make sense? > Depends on how focused one is on latency and what RTT differences the proxy is detecting. If you don't care about a few ms on some connections or the RTT variation between connections to a server is normally under 1ms then it can be better to use the memory and CPU cycles for other things like cache_mem. Amos _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
