A new method to implement AFij at core routers in diffserv networks

The most complicated part in DiffServ implementation belongs to AF BHP. Indeed, the complication of AF comes not only from its architecture, but also from its goals. This paper will propose a method to implement AFij of AF BHP at core routers in DiffServ networks. Using this method, we can make AF subclasses (AFij ) easily. Moreover, the performance of the system will be better since the mechanism has congestion control ability and fair sharing between subclasses in an AF class.
Nội dung trích xuất từ tài liệu:
A new method to implement AFij at core routers in diffserv networks’Tap ch´ Tin hoc v` Diˆu khiˆ n hoc, T.23, S.1 (2007), 71—79ıee.. a `.A NEW METHOD TO IMPLEMENT AFij AT CORE ROUTERSIN DIFFSERV NETWORKSLE HUU LAP1 , VU NHU LAN 2 , NGUYEN HONG SON31 ViceDirector of PTIT, Hanoi, Vietnam2 Institute3 Facultyof Information Technology Hanoi, Vietnamof Information Technology II PTIT Ho Chi Minh city branch, VietnamAbstract. Nowadays, the DiffServ architecture is considered the most prevalent solution of QoSprovision in IP networks. What is the best way to implement this architecture is still a questionwithout answer. The most complicated part in DiffServ implementation belongs to AF BHP . Indeed,the complication of AF comes not only from its architecture, but also from its goals. This paper willpropose a method to implement AFij of AF BHP at core routers in DiffServ networks. Using thismethod, we can make AF subclasses (AFij ) easily. Moreover, the performance of the system will bebetter since the mechanism has congestion control ability and fair sharing between subclasses in anAF class.´´´’’’aoaT´m t˘t. Hiˆn nay, kiˆn tr´c dich vu kh´c biˆt du.o.c xem nhu. mˆt giai ph´p dam bao chˆ t lu.o.ngoaeeu .ae......’ a o a o a ` e´´ e´´ e ’’oauedich vu trˆn mang Internet. C´ch thu.c thi tˆt nhˆ t kiˆn tr´c kiˆ u n`y c`n l` mˆt vˆ n dˆ dˆ ngo.ea.....´`` n ph´.c tap nhˆ t trong viˆc thu.c thi dich vu kh´c biˆt thuˆc vˆ AF BHP . T´ ph´.c tapınh u .aeaeo eTh`nh phˆaau .......´ e ´´ o’’ a a ` eacua AF khˆng chı l` vˆ n dˆ kiˆn tr´ c m` c`n l` muc tiˆu cua n´. B`i b´o dˆ xuˆ t mˆt phu.o.ng ph´poua o a .e ’ o a a `e a.’ .´’’’ .dˆ thu.c thi AFij cua AF BHP o. trong l˜i bˆ dinh tuyˆn trong mang dich vu kh´c biˆt. Su. dungeeaeo o ......’˜ aaa oo aeeephu.o.ng ph´p n`y, ta c´ thˆ chia l´.p AF th`nh c´c l´.p con AFij mˆt c´ch dˆ d`ng. Tuy nhiˆn, hiˆua ao eo..’ ´`´´´`’ e o’ a’ on˘ng cua hˆ thˆng s˜ tˆt ho.n khi co. chˆ c´ kha n˘ng diˆu khiˆ n t˘ c ngh˜n v` chia se cˆng b˘ ng gi˜.aae oe oee ae aau.c´c l´.p con trong l´.p AF .a oo1. INTRODUCTIONThe DiffServ architecture is based on a network model implemented over a complete Autonomous System (AS) or domain. All traffic entering and flowing through the domain aremanaged by clear and consistent rules. Depending on [2], IP packets entering network at edgerouters are classified and aggregated into different groups called Behavior Aggregates (BAs).Inside the domain, packets belonging to the same BA are forwarded according to previouslyestablished rules. Like this, the real work is creating classes of flows that travel through networks. Each flow is treated along the domain according to the class to which it belongs. Thus,the treatment of routers for classes is also established and takes forms called Per Hop Behaviors (P HB ). P HB are implemented by allocating resources quite differently inside routers.Currently, only two P HB are adopted and deployed, they are EF P HB and AF BHP . AFP HB is more intricate than EF P HB . Indeed, [3] specifies that AF P HB has four classes,72LAP LE HUU, LAN VU NHU, SON NGUYEN HONGcalled AF classes. Each AF class is further divided into three priority levels, which will becalled AF subclasses and denoted AFij . P HB s and AF classes are implemented by assigningresources of the system, such as buffer, bandwidth, service time, etc. There are numerous proposals and evaluations on implementing P HB for IP DiffServ Networks by various authors.For example, in [4] authors implemented EF P HB by using CBQ (Class Based Queuing) andproved that it is more efficient than priority scheduling or WRR (Weighted Round Robin) proposed by Christian Worm Mortensen. Some others, such as in [5] also implemented AF P HBby using CBQ, just called AFCBQ. Recently, in [6] used HTB (Hierarchical Token Bucket)proposed by Martin Devera to implement AF , and named AFHTB. [6] also used the packetdropping mechanism of RED for implementing AF subclasses. By this way, the packet dropping probability depends on the average queue length, qave , and with determined maximumthreshold/minimum threshold, the higher qave gets, the bigger the probability is. So, addingvarious values into qave will make different behaviors of the queue, which are truly AF subclasses. However, the method depends much on RED. It is nearly impossible to select suchRED parameters that the impact on network performance and on stability of the systemdoesn’t get worse. The proposed method in this paper bases on a controllable queue management (CQM) scheme instead of using RED, an active queue management (AQM) algorithm.The essence of the method is a mechanism, which includes controllers and token buckets asproposed in [1]. This mechanism also uses packet discarded levels to make difference of services (drop precedence) but these time token buckets are responsible for dropping packets inurgent situations.As mentioned above, the paper will focus on implementing AF subclasses in each AFclass. The implementation not only satisfies the requirements of AF class specification, butalso uses maximum capacity of the system without congestion and shares allocated resourcesof the system between subclasses in each AF class.The rest of the paper is structured as follows. In section II, we present the principle ofimplementation and operation inside the mechanism. This section not only explains how tomake various drop precedences (AFij ), but also guides to share resources between them. Inthe next section, we clarify the method by establishing the dynamics of the mechanism. Insection IV, we simply validate the proposed method by a computer simulation. The resultsof simulation also show that the system is always protected from congestion and can obtainthe highest utility. In the final section, we present our conclu ...