Pull-Based Load Distribution Among Heterogeneous Parallel Servers

#Networks #Communications #Information #Theory
Share

Prof. Sasha Stolyar, Lehigh University, will give a presentation on pull-based load distribution among heterogeneous servers, and in particular, he will focus on the case of multiple routers. The presentation will be held at Bell Laboratories, Nokia, in Room 6A-209. Please register before the meeting, as one would have to be signed in to get to the meeting room.



  Date and Time

  Location

  Hosts

  Registration



  • Date: 20 Jul 2016
  • Time: 02:00 PM to 03:00 PM
  • All times are (UTC-04:00) Eastern Time (US & Canada)
  • Add_To_Calendar_icon Add Event to Calendar
  • Bell Laboratories, Nokia
  • 600 Mountain Ave
  • Murray Hill, New Jersey
  • United States 07974

  • Contact Event Host
  • Starts 05 July 2016 12:00 AM
  • Ends 20 July 2016 01:00 PM
  • All times are (UTC-04:00) Eastern Time (US & Canada)
  • No Admission Charge


  Speakers

Prof. Sasha Stolyar

Topic:

Pull-Based Load Distribution among Heterogeneous Parallel Servers - the Case of Multiple Routers

Load distribution in modern cloud-based data processing systems is a challenging problem, due in particular to their very large scale. Algorithms are needed that, in addition to good performance (e.g., small waiting times and/or blocking), are efficient in the sense of low rate of signaling/communication between system components. Motivated by this
   general problem, we consider a heterogeneous service system, consisting of several (different) large server pools, and one or more routers. Each router receives an equal fraction of the customer arrival flow, and has to route each customer to a server immediately upon arrival. We cosider the asymptotic regime in which the total arrival rate and pool sizes scale to infinity simultaneously, while the system load remains subcritical. We study a 'pull-based' routing scheme (labeled PULL) and prove that, under this scheme and certain assumptions, both waiting times and blocking probabilities asymptotically vanish. In particular, this shows that the performance of PULL is much superior to that of the celebrated 'power-of-d-choices' routing algorithm, and this is achieved with much lower signaling overhead.

Address:200 West Packer Avenue, , Bethlehem, United States, 18015

Prof. Sasha Stolyar

Topic:

Pull-Based Load Distribution among Heterogeneous Parallel Servers - the Case of Multiple Routers

Biography:

Address:Bethlehem, United States