> > 1) Predictive, proactive app scaling that learns load patterns and adjusts > instance/resource minimums on a schedule. This could be in lieu of manually > specifying cpu / memory requirements (assuming controls to prevent runaway > scaling.) > Tobi Knaup built an nifty POC that scales apps based on SLA-y stuff like latency. It's not quite what you're looking for, but it's a practical pattern available now.
As for removing the specification of cpu/memory requirements, Christina Delimitrou gave a Twitter talk on exactly this: https://www.youtube.com/watch?v=YpmElyi94AA (Note it's still in the research stage) > 3) Cloud-specific VM management that attempts to maintain a constant > level of excess capacity for scaling operations (i.e. always maintain 20% > open slave capacity for scaling up, and bin pack for scaling down when > possible) > I'm hacking on a tool to manage autoscaling of Mesos slaves. First iteration is pluggable with EC2 ASG support out of the box. I'll have something public next week. Haven't started thinking about task relocation or bin packing, though. Hopefully smarter folks than me will help out on that side :) 4) Dynamic HAProxy load balancing (updates nodes in real-time) w/ public > IP/DNS bindings. > I can only speak to the Marathon space: - https://github.com/mesosphere/marathon/blob/master/bin/haproxy-marathon-bridge - https://github.com/Wizcorp/frontrunner - https://github.com/opencredo/mesos_service_discovery If you want to roll your own, see: https://groups.google.com/forum/#!topic/marathon-framework/EDzZ6gLHd3Q If you're on AWS, I'll be releasing something potentially useful (and CloudFormation-based) next week. On Fri, Aug 22, 2014 at 1:01 PM, Jeremy Jongsma <jer...@barchart.com> wrote: > Hi all - I just got back from MesosCon and am excited about doing a deeper > evaluation of Mesos in our environment. There are a few areas I'm very > interested in and could contribute some development effort towards. What > work is being done in these areas (if any?) Sorry if these have been talked > to death already: > > 1) Predictive, proactive app scaling that learns load patterns and adjusts > instance/resource minimums on a schedule. This could be in lieu of manually > specifying cpu / memory requirements (assuming controls to prevent runaway > scaling.) > > 2) Cassandra framework. We've done a lot of work scripting multi-region > Cassandra cluster installs, I can probably contribute something here. > > 3) Cloud-specific VM management that attempts to maintain a constant level > of excess capacity for scaling operations (i.e. always maintain 20% open > slave capacity for scaling up, and bin pack for scaling down when possible) > > 4) Dynamic HAProxy load balancing (updates nodes in real-time) w/ public > IP/DNS bindings. > > Looking forward to getting more involved! > > Jeremy >
