The scheduler, which lives inside the ResourceManager, is responsible for granting containers (each which has a requested memory capability) to applications and choosing where to put them. It makes no attempt co-locate containers, but Application Masters, which are responsible for requesting resources, can request resources on particular nodes or racks. Does that help to answer your question? I'm not super familiar with the OS literature on this.
All memory management on a single node is done by the OS. -Sandy On Thu, Jul 18, 2013 at 3:03 AM, 牛兆捷 <[email protected]> wrote: > For a distributed sense, such as whole memory management for the resource > manager. > > btw,what is it for a single node? > > 2013/7/18 Sandy Ryza <[email protected]> > > > Hi Zhaojie, > > > > Do you mean on each node, or in a distributed sense? > > > > -Sandy > > > > > > On Wed, Jul 17, 2013 at 5:56 PM, 牛兆捷 <[email protected]> wrote: > > > > > Hi All: > > > > > > What is the memory management algorithm in Yarn? > > > > > > Is it buddy policy? > > > > > > -- > > > *Sincerely,* > > > *Zhaojie* > > > * > > > * > > > > > > > > > -- > *Sincerely,* > *Zhaojie* > * > * >
