Paper review: A Framework for Receiver Oriented Differentiated Services

Reviewer: Kevin Hofstra

  1. Differentiated services is a way of providing scalable QoS assurances on the internet.  The current services operate exclusively from the sender (RODS).  Is there a way of having the services selected by the receiver (RODS)?  Will this create additional overhead or latency?
  2. This paper discusses the implementation of Receiver Oriented Differentiated Services.
  3.  

A.     There are 3 types of Differentiated services:

i.                     Open contract mode (OC)  (highest level always chosen)

ii.                   Predefined service mode (PDS)  (agreement is defined)

iii.                  On request mode (OR)  (requested and met if possible)

B.     RODS inherits the basic architecture of DiffServ.  RODS and SODS are meant to co-exist and compliment each other according to the type of application.  RODS is based on the credit scheme between the sending and receiving Bandwidth Broker (BB) using SLAs.  In a system that already has SODS, RODS does not produce very much additional overhead.  Most of its issues are associated with the credit negotiations of the end receiver and his BB. 

C.       The question of whether a packet should use RODS or SODS when both are specified is only briefly mentioned and is going to cause some conflicts.

D.     The extra overhead and latency of establishing a SLA between the sender and a specifying receiver may actually be costly if it is a short one time transfer.

  1. Critique the main contribution
    • Significance- 4  I believe that they have come up with some very important functionalities for some of today’s internet applications.  Many applications would rather have the receiver specify the QoS.  This may very well be the answer to bi-directional differentiated services.  The actual overhead in terms of implement overhead is smaller because it uses the existing SODS framework which would ease its deployment tremendously.
    • Convincing- 2  I think that is article is somewhat misleading.  The figures fail to show the initial increase in latency in additional RODS negotiation which would be compounded in a short transfer.  They have shown the drastic increase in number of successful connections but do not give any indications how they received this performance gain.  The performance is also based on a system where SLAs had already been negotiated and were not very complex.
  2. System researchers and builders should recognize that there would be additional overhead in trying to keep the SLAs of the receivers current and accessible.  The contract between the receiver and all possible senders must be specified for all possible incoming packets which would create significant additional negations with the BB.  The bandwidth broker will not only have the additional overhead of keeping track of the SLAs of each of its receivers using RODS, but will also have to do all the negotiations with neighboring domains.  This paper does a great job of showing how the performance will be increased in a small network with established contracts and SLA’s.  However in the internet with constantly changing contracts, many more intermediate domains, large numbers of possible senders and receivers, I believe that it will be difficult for the BB to adequately achieve stable SLA’s.  There is not additional overhead per session, but much more additional overhead to negotiate the initial session.