[squid-users] [ext] Netdb. state too big

Amos Jeffries squid3 at treenet.co.nz
Wed Jan 9 13:35:19 UTC 2019


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


More information about the squid-users mailing list