Using WUGS Kits to Improve the Quality of Service Min Song, Old - PowerPoint PPT Presentation

Using WUGS Kits to Improve the Quality of Service Min Song, Old Dominion University Mansoor Alam, University of Toledo

Outline • What we have done • Teaching • Course Projects • Research • Theses • What we plan to do • QoS Control in VPNs • Evolutionary Scheduling Implementation • AAL5 Cell Encryption and WFQ on MSR

Course Projects (1) EECS 6160 - B-ISDN and ATM Networks Project 1: Install NetBSD, WUGS programs[1] and build a simple switched LAN R1 S1 WUGS #2 6 3 R2 S2 4 5 6 3 4 CP2 0 CP1 WUGS #1 APIC Project 2: Configure WUGS to route cells from the senders to the receivers • A single AAL0_cell • A frame of AAL0 cells • AAL5 cells

Course Projects (2) Project 3: Install SPCs on WUGS, and download the kernel and applications server client Project 4: Run the applications on SPC R1 S1 WUGS #2 6 3 R2 S2 4 5 6 3 4 CP2 0 CP1 WUGS #1 APIC

Three-phased Weighted Fair Queueing (TWFQ)[6] • By separating the traffic into three groups and having a dedicated server for each group can significantly reduce the packet latency. Master Controller ….. group 1 Server 1 ….. group 2 Server 2 ….. group 3 Server 3 • Weighted Fair Queuing based on Flow-timestamps (WFQF) is used in each of the three servers. • TWFQ has been Implemented on SPC

WFQ based on Flow-timestamps (WFQF)[4] 1. On arrival from the input link, a packet p is classified, time stamped, and stored in its dedicated queue; 2. If this queue is inactive, packet p is time stamped with the start tag, computed as = + i , 1 i , 1 i , 1 S ( q ) max{ V [ A ( q . p )], S ( q )} x , y x , y x , y 1 i: input port # φ i , 1 i , 1 ( A ( q . p )) x , y x , y x: group # y: flow ID V[.] is the smallest start tag over all active queues within the corresponding group. If this queue is active, the start tag is not updated; 3. Among the HOL packets, each server picks up the one with the smallest start tag; 4. After packet p is scheduled and forwarded, if this queue remains active, its queue timestamp is updated as follows: 1 = + i , 1 i , 1 S ( q ) S ( q ) φ , , x y x y i , 1 i , 1 ( A ( q . p )) x , y x , y

Evolutionary Scheduling[8] 2 8 − ∑ 2 1 Evolutionary O i S f d ( O , S ) = 10 -4 i Scheduling 2 = i 1 S 1 : the available bandwidth resource O 1 : the desire for the bandwidth S 2 : the maximum delay a cell of a particular flow can wait in a queue Group Group Group O 2 : the desire for lesser delay 1 2 3 The higher the fitness value of a queue, the stronger is the queue in competition with others q1 q2 q4 q5 q6 q3 To avoid the group/queue starvation, guarantee value is used. The guarantee value is used to prevent starvation of a resource. During the selection of a winner group or a winner queue, the corresponding group or queue guarantee value is decreased. We set the initial guarantee value as follows: g = ( f d ( O , S ) >> 5) +1, such that 1 ≤ g ≤ 8 for 0 ≤ f d ( O , S ) ≤ 255

i LPF Algorithm[3,5] N = ∑ N = ∑ R i ( n ) C j ( n ) L ( n ) L ( n ) , i , j i , j j i Input Port Occupancy Output Port Occupancy L ( n ) : queue length at time n i , j (1,1) 1 1 (1,2) i LPF favors queues with (2,1) 2 high port occupancy 2 (2,2) Step 1: Sorting 1. Sort inputs and outputs based on their port occupancies 2. Reorder requests according to their input and output occupancies Step 2: Matching 1. for each output from largest to smallest 2. for each input from largest to smallest 3. if (there is a request) and (both input and output unmatched) 4. then match them

i LPF: Pros & Cons Cons Pros • 100% throughput • Link-dependent ∑ < ∞ E [ L ( n )] • • Non-starvation-free , i j i , j = l l 10 Hot-spot 11 11 l = l 6 12 12 = l 12 21 l 21 = l 2 l 22 22

Worst-case Controller (WC)[4,7] 1 worst-case window ∆ = − − l ( t ) [ l ( t ) l ( t nl )] i , j i , j i , j n ∆ − = ∆ − ∆ − l ( t , t nl ) l ( t ) l ( t nl ) t-nl … t-2l t-l t t+l i , j i , j i , j nl d = ∆ + ∆ + ∆ − m p [ l ( t ) l ( t ) l ( t , t nl )] i , j i , j i , j i , j x nl ≥ = if {( m ( t ) M ) and ( WC _ match 0 )} i , j i , j then fill in i to port j’s worst-case matching list WL j • Can be easily incorporated with the existing algorithms • To be Implemented on SPC/FPX

Worst-case LPF (WLPF) Step I: Sorting Sort inputs and outputs based on their occupancies Reorder requests according to their input and output occupancies Step II: Worst-case matching for (each output from 1 to N ) for (the worst-case list from left to right) if (both the output and the input unmatched) , = then match them, WC _ match 1 i j Step III: i LPF and final matching for each output from largest to smallest for each input from largest to smallest , = if ( ) WC _ match 1 i j , = then match them, LPF _ match 1 i j if (both input and output unmatched) , = then match them, LPF _ match 1 i j

Outline • What we have done • Teaching • Course Projects • Research • Theses • What we plan to do • QoS Control in VPNs • Evolutionary Scheduling • AAL5 Cell Encryption and WFQ on MSR

QoS Control in VPNs [9] UT • Security and quality of service are essential for the network to satisfy the needs of business communication. • A well designed VPN offers a public network the same security as a private network. The fundamental VPN technologies are generally E deployed at the end systems. x t r a ODU n e t Site 2 • What happens between the VPN gateways V P N N P V t e n a r t n I ODU Site 1 • In order to provide QoS guarantees in the Internet, all packets of a flow must traverse the same path, and some means for reserving resources along that path must exist. • We plan to build two sites at ODU and create two VPNs. MPLS technology and the presented algorithms will be implemented on WUGSs.

Evolutionary Scheduling • The evolutionary scheduling algorithm described early was just tested by simulation using simulator ATMSIM [2] • We plan to design the fitness function including more parameters • Computation complexity • Implementation SPC/FPX • Applications

Encryption and WFQ • AAL5 cell encryption on SPC/FPX (David) • Weighted Fair Queuing in MSR (Sachin)

Part of References [1] http://www.arl.wustl.edu/gigabitkits/kits.html [2] Min Song, Mansoor Alam, and Uday Jarajapu, “High Performance ATM Network Simulators”, published in proceedings of International Society for Computers and Their Applications 17th International Conference on Computers and Their Applications, April 2002, San Francisco. [3] A. Mekkittikul, N. McKeown, “A Starvation-free Algorithm For Achieving 100% Throughput in an Input-Queued Switch”, Proc. of the IEEE International Conference on Communication Networks, 1996. [4] Min Song, “Design and Performance Analysis of Efficient Packet Scheduling Algorithms for Internet Routing Switches”, PhD dissertation, The University of Toledo, 2001. [5] A. Mekkittikul and N. McKeown, “A Practical Scheduling Algorithm to Achieve 100% Throughput in Input-Queued Switches”, Proceedings of the IEEE INFOCOM, 1998. [6] Mansoor Alam, Qing Tan, Min Song, and X. Hao, “QoS Control and Performance Analysis of a Gigabit Network”, published in proceedings of International Society for Computers and Their Applications 17th International Conference on Computers and Their Applications, April 2002, San Francisco. [7] Min Song, Mansoor Alam, “Two Scheduling Algorithms for Input-queued Switches Guaranteeing Voice QoS”, published in proceedings of IEEE GLOBECOM 2001, Texas, 2001. [8] Min Song, Mansoor Alam, and Sachin Shetty, “Evolutionary Computing in a Multi-Service Routing Switch Packet Scheduling”, published in proceeding of the 17th International Conference on Computers and Their Applications, April 2002. [9] Min Song, Mansoor Alam, research proposal submitted to NSF, 2002.

Thank You!