On Sep 16, 2014, at 7:20 PM, ilya musayev <ilya.mailing.li...@gmail.com> wrote:
> Hi all, > > Would you know where we stand with Mesos and Docker? > That's a big question. Mesos is a resource allocator that multiple frameworks can use to run workloads of various sorts. The interest is to mix workloads: big data, long running services, parallel computing, docker in order to maximize utilization of your resources. For instance Aurora (mesos framework) can execute long running services within docker containers. The challenge with docker is the coordination of multiple containers. Kubernetes for example coordinate docker containers to run HA applications. What we see (IMHO) is things like Kubernetes being deployed in the cloud (gce, azure, backspace are currently "supported" in kubernetes). And at mesoscon, there was a small demo of running kuberneters as a mesos framework. So…bottom line for me is that I see Mesos and everything on top as a workload that can be run in CloudStack. Similar thing with CoreOS. If a CloudStack cloud makes available CoreOS templates, then users can start CoreOS cluster and manage Docker straight up or via Kubernetes (because of course there is CoreOS "support" in Kubernetes). Hence, there is nothing to do, except for CloudStack clouds to show that they can offer Mesos* or Kubernetes* on demand. However if we were to re-architect CloudStack entirely, we could use Mesos as a base resource allocator and write a VM framework. The framework would ask for "hypervisors" to mesos and once allocated CloudStack would start them…etc. The issue would still be in the networking. The advantage is that a user could run a Mesos cluster and mix workloads: CloudStack + Big Data + docker…. Anything we can do to make CoreOS "cloud stackable" and create a cloudstack driver in Kubernetes would be really nice. > Thanks > ilya
