>> I've known about this (nginx getting slower over time) but not had time to >> investigate thoroughly. I think its an nginx-specific issue; it may be to do with the fact that we're running in the special "non-forking” mode (master_process off), or not.
One interesting point here is the Nginx doesn’t start at the expected performance level (approx 5,500 transactions), it starts at approx 1,300 and then slows down. Whereas it appears from your mathopd test that mathopd starts at a high transaction rate - assuming of course that the transaction rate is similar for mathopd when running in a non-rump environment on the same machine - as Antii has enquired about. as
