A while ago me and Singer Wang (another DBA here at pythian) was working on a problem where the slave went from 800M to 1200 to 2700 to 3200 to ~800M again. I have to admit I haven't seen this before.The specific setting was 5.0.45 slave that also acted as a master, and the instance the experienced the problems was running 5.0.50.After upgrading to 5.0.74, it worked again. But here's a few of the show slave status while experiencing these problems: