QoS Strategy in DiffServ aware MPLS environment Teerapat Sanguankotchakorn, D.Eng. Telecommunications Program, School of Advanced Technologies Asian Institute of Technology P.O.Box 4, Klong Luang, Pathumthani, 1212, Thailand E-mail: teerapat@ait.ac.th
Service model how capacity is shared Traffic engineering how much capacity routing,... Quality of service transparency throughput accessibility Technology filtering, scheduling queue management etwork business model congestion pricing volume pricing
Differentiated Services (DiffServ) Required Mechanisms for Service Differentiation - Classification: To identify the aggregation (class) to which a packet belongs based on their different QoS requirements. Scheduling: To determine which packet will be transmitted according to its priority specified by DS Code Point. Queue Management: To decide which packets will be dropped, when the buffer overflows in the case of network congestions. Incoming Packets classifier Marker Meter Conditioner Outgoing Packets Principle of Service Differentiation
Multi Protocol Label Switching (MPLS) Fast routing Provide bandwidth management Less load on core routers Use of labels in packet headers (short, fixed length and locally significant) Traffic engineering (classification and identification of IP packets with a label and forwarding the packets to a switch or router that is modified to operate with such labels. Signaling protocols used in MPLS (Label Distribution Protocol (LDP)) o end-host protocol component Quality of Service (QoS)
DiffServ + MPLS A scheme which is mutually beneficial for both MPLS provides DiffServ with Path protection and restoration. DiffServ acts as CoS architecture for MPLS SO, MPLS with DiffServ can give network designers the flexibility to provide different treatment to certain QoS classes that need path-protection.
Domain Model LSP Diffserv aware MPLS Domain LER LER Some kind of mapping is required to convert IP header values to label LSR LSR LSR LSR LER LER LER: Label Edge Router LSR: Label Switch Router LSP: Label Switch Path
Why: Mapping DiffServ to MPLS Label Switch Routers (LSRs), don t see the IP header and DSCPs in ToS field of IPv4 header. LSR only reads the Label contents and decides the next hop How: The contents of 6 bits DSCP is mapped into 3 bits EXP field of Label There are two options to map DSCP value into the Label E-LSP EXP-inferred-PSC LSP PHB is determined from EXP bits o additional signaling is required EXP-PHB mapping is configured Shim header is required Up to 8 PHBs per LSP With Bandwidth reservations, the bandwidth is shared by set of transported PSCs L-LSP Label-only-inferred-PSC PHB is determined from Label or from label and EXP/CLP bits PHB or PSC is signaled at LSP setup Label-PHB mapping is signaled. EXP/CLP-PHB mapping is well known Shim or link layer header may be used One PHB per LSP except for AF and PSC per LSP for AF With bandwidth reservation, the bandwidth is per-psc
Mapping Traffic to LSPs Incoming Packets Ingress ode With both Diffserv and MPLS enable Premium Traffic High Priority Traffic LSP 1 LSP 2 A network may have multiple classes of traffic. For a same destination, there might be different classes. ow, we can map different classes to different LSPs. Problem: Best Effort Traffic LSP 3 Again the delay and jitter is possible as all the premium traffic is following the same physical path. We can solve the above mentioned problem by splitting the traffic into different LSPs, even the destination is the same. We can allocate certain Bandwidth for each service class in a single LSP.
Methodology Calculation of: Bandwidth Utilization per traffic flow Throughput ( T stop r r P T Ps 8bits t 1 s start 8bits ) 1 where; where; P s r t T T start stop umber of packets received at destination router in bytes Packet size Time in second Start time of each traffic flow in second Stop time of each traffic flow in second Dropped packets Mean End-to-End Delay g g i1 r D r 1% i where; where; g r Di Ts i Td i umber of packets generated at source umber of packets received at source End-to-End delay of packet i = Ts i Td i Time of packet i en-queue at source router Time of packet i receive at destination
DiffServ + MPLS Application Scenario Ingress ode Premium Traffic LSP 1 Incoming Packets With both DiffServ and MPLS enable High Priority Traffic LSP 2 Best Effort Traffic LSP 3
etwork Topology: (Simple) Overall Packet Loss in %age SRC 1 SRC 2 SRC 3 For Simple etwork e1 c2 5Mbit/s c1 c3 5Mbit/s e2 DEST 1 DEST 2 DEST 3 %age Packet Loss 5 4 3 2 1-1 1 2 3 4 5 6 rate 4 in Mbps Average End to End Delay SRC 4 3Mbit/s 3Mbit/s c4 Domain 3Mbit/s c5 DEST 4 delay in msec.25.2.15.1.5 1 2 3 4 5 6 rate4 in Mbps Traffic sources are CBR sources on UDP agents. All packets follows the same smallest path e1-c1-e2 for simple case. For each case we will keep constant all the above 3 sources and will vary only last source from 5kbps to 5kbps. Other constant rates are: Rate1: 1.9 Mbps (Real Time 1---RT1) Rate2: 1.1 Mbps (Real Time 2---RT2) Rate3: 1. Mbps (High Priority Best Effort---HPBE) Rate4: 5kbps 5Mbps (Simple Best Effort SBE) Simulation time is 3. seconds Link Utilization in %age 25 2 15 1 5 Overall Link Utilization of etwork 2 4 6 rate4 in Mbps
etwork Topology: (DiffServ) Overall Packet Loss in %age SRC 1 For Diffserv etwork C2 C3 DEST 1 %age Packet Loss 5 4 3 2 1-1 1 2 3 4 5 6 rate 4 in Mbps SRC 2 SRC 3 SRC 4 E1 5Mbit/s C4 C1 DiffServ Domain C5 5Mbit/s 3Mbit/s 3Mbit/s 3Mbit/s E2 DEST 2 DEST 3 DEST 4 delay in msec.2.15.1.5 Average End to End Delay 1 2 3 4 5 6 rate4 in Mbps Traffic sources are CBR sources on UDP agents For each case we will keep constant all the above 3 sources and will vary only last source from 5kbps to 5kbps. Other constant rates are: Rate1: 1.9 Mbps (Real Time 1---RT1) Rate2: 1.1 Mbps (Real Time 2---RT2) Rate3: 1. Mbps (High Priority Best Effort---HPBE) Rate4: 5kbps 5Mbps (Simple Best Effort SBE) Simulation time is 3. seconds Link Utilization in %age 25 2 15 1 5 Overall Link Utilization of etwork 2 4 6 rate4 in Mbps
etwork Topology: (MPLS) Overall Packet Loss in %age For MPLS etwork SRC SRC 1 LSR4 SRC 2 LSR7 LSR8 5Mbit/s LSR5 5Mbit/s LSR6 DEST 11 DEST 12 DEST 13 %age Packet Loss 4 3 2 1-1 1 2 3 4 5 6 rate 4 in Mbps Average End to End Delay SRC 3 3Mbit/s 3Mbit/s LSR9 3Mbit/s MPLS Domain LSR1 DEST 14 If no ER-LSP is defined, all packets follow the same smallest path LSR4-LSR5-LSR6. In our case, we establish ER-LSP as soon as simulation starts. Traffic sources are CBR sources on UDP agents For each case we will keep constant all the above 3 sources and will vary only last source from 5kbps to 5kbps. Other constant rates are: Rate1: 1.9 Mbps (Real Time 1---RT1) Rate2: 1.1 Mbps (Real Time 2---RT2) Rate3: 1. Mbps (High Priority Best Effort---HPBE) Rate4: 5kbps 5Mbps (Simple Best Effort SBE) Simulation time is 3. sec delay in msec Link Utilization in %age.2.15.1.5 8 6 4 2 1 2 3 4 5 6 rate4 in Mbps Overall Link Utilization of etwork 5 1 15 rate4 in Mbps
Conclusions From our Dumbbell network topology, The overall performance (packet Loss, End-to-End elay, Link utilization) of DiffServ is quite similar to the Simple scheme, but MPLS is superior than DiffServ. In Simple case, there is no mechanism to classify the packets, so packets are randomly dropped from all 4 sources. In DiffServ, the low priority packets (Best effort packets) are discarded. Since RED queue has been used, some early dropping of high priority traffic can also be predicted in the network. If we simply assign the priority in descending order to RT1,RT2 then HPBE and SBE in case of MPLS, Link utilization can be increased by mapping different traffic through different routers.
Thank You!