Roy M. wrote:
According to 3.0 manual: ========= weighted-round-robin to define a set of parents which should be used in a round-robin fashion with the frequency of each parent being based on the round trip time. Closer parents are used more often. ========= Currently I have 3 Apache web servers, A, B, C, where B has Dual CPU and more memory, they are under the same private network. So I assign B with more weight. #A parent 80 0 no-query originserver weighted-round-robin login=PASS weight=1 #B parent 80 0 no-query originserver weighted-round-robin login=PASS weight=2 #C parent 80 0 no-query originserver weighted-round-robin login=PASS weight=1 However, from the access log from these 3 web servers, I found that the MISS request to #B is only around 130% higher than A and C Is it normal or I misunderstood the weighted-round-robin settings?
Small mis-reading perhapse. weighted-round-robin works the same as round-robin only its based on the peer RTT (network delay to reach peer).
In vanilla robin, the counter gets 1 added, which evenly balances requests through the peers, regardless of network trouble or anything else.
In weighted, the counter gets RTT/weight, which balances things more in favour of close peers. But weight= can give an extra boost to preferred peers or a manual balancing if the expected RTT (in ms) to that peer is large. The division is never allowed to produce a non-integer or stat under 1.
Amos -- Please use Squid 2.7.STABLE1 or 3.0.STABLE6
