On Tue, Sep 15, 2020 at 09:24:32AM +0200, Willy Tarreau wrote: (...) > There's no such ongoing work that I'm aware of but that has always > been a subject of interest to me (I even wrote down the algorithm to > compute weights by measured response times using a low-pass filter a > decade ago but I lost my notes and never felt like doing work again). > So if anyone is interested in this subject, we can continue this > conversation till we reach something that looks like a possible design > roadmap.
By the way there is another aspect that has always interested me around this, which is that it would simplify the confuguration of a server's maxconn when it's shared between multiple backends or even multiple LB nodes. It will certainly lead to oscillations sometimes but that's manageable by using randomness in increments so that multiple LB nodes cannot enter into resonance. Willy
