Cos, That is pretty apparent at first blush, but from a high enough level one can say I want to run Spark on Mesos vs. Spark on YARN and what are the differences in (isolation, performance, management, extensibility, accounting...) and how does one quantify those differences if there are any.?.?.?
Cheers, Tim ----- Original Message ----- > From: "Konstantin Boudnik" <[email protected]> > To: [email protected] > Sent: Monday, July 8, 2013 12:58:20 AM > Subject: Re: Mesos vs. YARN > > Tim. > > While feature comparison is in interesting point indeed, I'd say that this is > a little bit like comparing oranges to apples. Leaving out things like > implementation, foundation technologies, performance guarantees, etc. I'd > like > to point to one major difference between the two: > - YARN is Hadoop-centric resource manager, that relies on parts of Hadoop > implementation, e.g. some parts of o.a.h.fs, o.a.h.io, o.a.h.security, > etc. I believe it is focused on deployment of applications that either > are fully Hadoop specific (e.g. MR) or dependent on HDFS (e,g. HBase as > in project HOYA) > - Mesos is application agnostic, e.g. you can deploy Hadoop with it; or > Spark; or else. > And that, IMO, the most important difference of all. > > Regards, > Cos > > On Fri, Jul 05, 2013 at 05:44PM, Tim St Clair wrote: > > Folks - > > > > With YARN reaching beta stage, has anyone done a full feature > > comparison between YARN and Mesos? (last email thread ~ 2 years old). > > > > Also, are there any performance benchmarks between the two for comparable > > operations. > > > > Cheers, > > Tim >
