2 Replies Latest reply on Jun 22, 2010 4:07 PM by mikikian

    Scalability of Stratus

    mikikian Level 1

      Does Adobe have a white paper or test results that can prove its claim that Stratus could "scale to millions or billions of people":


      RTMFP groups is more efficient now than the original rendezvous service. RTMFP groups allow developers to scale to millions and billions of people worldwide because it doesn’t source everything from a single peer or server cluster.

       

      I was just wondering if this was actually tested/proven or not.

       

      Thanks in advance!

        • 1. Re: Scalability of Stratus
          Michael Thornburgh Adobe Employee

          > Does Adobe have a white paper or test results that can prove its claim that Stratus could "scale to millions or billions of people"

           

          we do not have a white paper or test results that can prove this assertion.

           

          furthermore, "Stratus" is the name of Adobe's hosted RTMFP rendezvous service.  "RTMFP Groups" is the scalable, self-organizing P2P mesh technology that is included in Flash Player 10.1.  i assume your question really is "can we prove that RTMFP Groups can scale to millions or billions of people", not "can the Stratus service handle millions or billions of simultaneous users".  we would have to buy additional servers to handle "millions or billions" of simultaneous users on our Stratus cluster, although we're in a good position to handle near-term anticipated growth with the release of Flash Player 10.1.

           

          members of an RTMFP Group maintain O(log n) direct connections to other members (approximately 2*log n + 13 direct connections).  so in a group with 1000 members, each member would have about 33 direct connections.  for a group with 1,000,000 members, each member would have about 53 direct connections.  for a group with 1,000,000,000 members, each member would have about 73 direct connections.  the P2P multicast diameter of a group with default settings is approximately log4 n.  we believe scaling with O(log n) is "scalable".  we have tested groups with tens of thousands of members in the lab.

          • 2. Re: Scalability of Stratus
            mikikian Level 1

            Thanks Michael!

             

            You correctly answered the question I wanted to ask