I honestly think it solves a different problem. Apple runs Siri on the biggest mesos cluster in the world (pretty much verbatim quote from MesosCon 2015. I was

I honestly think it solves a different problem. Apple runs Siri on the biggest mesos cluster in the world (pretty much verbatim quote from MesosCon 2015. I was there). Twitter runs clusters over 10k physical nodes. Swarm and compose are cute, but they simply aren't made for that kind of massive scale. The inverse of course, is that they are incredibly easy to setup, and they work for a very large number of workloads reasonably well. I manage mesos clusters as part of $day_job. There is a certain level of setup required and if you don't need the scale, you simply don't need mesos for it. I really see them as complementary in the longer term. Perhaps a swarm mesos framework could be written. It is highly unlikely that Docker will ever do the resource scheduling better than Mesos.