Performance Measurements of MPLS Traffic Engineering and QoS

By Tamrat Bayle Reiji Aibara Kouji Nishimura

Multiprotocol Label Switching

Traditional IP Routing Disadvantages Need for MPLS MPLS basics and terminologies Experiments

Traditional IP Routing

Choosing the next hop

Open Shortest Path First (OSPF) to populate the routing table Route look up based on the IP address Find the next router to which the packet has to be sent Replace the layer 2 address

Each router performs these steps

Traditional IP Routing (contd)

Distributing Routing Information

Address Prefix Address Prefix Path Address Prefix Path

Path

125.50

125.5 0 145.40

3 3

125.50 145.40

0 1

0
3

125.5 20

You can reach 125.50 through me

1

You can reach 125.50 and 145.40 through me You can reach 145.40 through me

145.40

Distributing Routing Information(contd)

Address Prefix Address Prefix Path Address Prefix Path

Path

125.50

125.5 0 145.40

3 3

125.50 145.40

0 1

0
3

125.5 20

Data
Data 125.50.33.85
1

125.50.33.85
145.40

Disadvantages

Header analysis performed at each hop Increased demand on routers Utilizes the best available path Some congested links and some underutilized links!
Degradation of throughput Long delays More losses

No QoS
No service differentiation Not possible with connectionless protocols

Need for MPLS

Rapid growth of Internet New latency dependent applications Quality of Service (QoS)
Less time at the routers

Traffic Engineering
Flexibility in routing packets

Connection-oriented forwarding techniques with connectionless IP

Utilizes the IP header information to maintain interoperability with IP based networks Decides on the path of a packet before sending it

What is MPLS?

Multi Protocol supports protocols even other than IP

Supports IPv4, IPv6, IPX, AppleTalk at the network layer Supports Ethernet, Token Ring, FDDI, ATM, Frame Relay, PPP at the link layer

Label short fixed length identifier to determine a route

Labels are added to the top of the IP packet Labels are assigned when the packet enters the MPLS domain

Switching forwarding a packet

Packets are forwarded based on the label value NOT on the basis of IP header information

MPLS Background

Integration of layer 2 and layer 3

Simplified connection-oriented forwarding of layer 2 Flexibility and scalability of layer 3 routing

MPLS does not replace IP; it supplements IP Traffic can be marked, classified and explicitly routed QoS can be achieved through MPLS

IP/MPLS comparison

Routing decisions
IP routing based on destination IP address Label switching based on labels

Entire IP header analysis

IP routing performed at each hop of the packets path in the network Label switching performed only at the ingress router

Support for unicast and multicast data

IP routing requires special multicast routing and forwarding algorithms Label switching requires only one forwarding algorithm

Key Acronyms

MPLS MultiProtocol Label Switching FEC Forward Equivalence Class LER Label Edge Router LSR Label Switching Router LIB Label Information Base LSP Label Switched Path LDP Label Distribution Protocol

Forwarding Equivalence Class (FEC)

A group of packets that require the same forwarding treatment across the same path Packets are grouped based on any of the following
Address prefix Host address Quality of Service (QoS)

FEC is encoded as the label

FEC example

Assume packets have the destination address as 124.48.45.20 143.67.25.77 143.67.84.22 124.48.66.90
FEC 1 label x FEC 2 label y

143.67.25.77 143.67.84.22

124.48.45.20 124.48.66.90

FEC example (contd)

- Assume packets have the destination address and QoS requirements as 124.48.45.20 qos = 1 143.67.25.77 qos = 1 143.67.84.22 qos = 3 124.48.66.90 qos = 4 143.67.12.01 qos = 3
FEC 1 label a FEC 2 label b FEC 3 label c FEC 4 label d

143.67.25.77

124.48.45.20

143.67.84.22 143.67.12.01

124.48.66.90

Label Edge Router (LER)

Can be an ATM switch or a router Ingress LER performs the following:

Receives the packet Adds label Forwards the packet into the MPLS domain

Egress LER removes the label and delivers the packet

LER

Label Switching Router (LSR)

A router/switch that supports MPLS Can be a router Can be an ATM switch + label switch controller Label swapping
Each LSR examines the label on top of the stack Uses the Label Information Base (LIB) to decide the outgoing path and the outgoing label Removes the old label and attaches the new label Forwards the packet on the predetermined path

Label Switching Router (contd)

Upstream Router (Ru) router that sends packets Downstream Router(Rd) router that receives packets
Need not be an end router Rd for one link can be the Ru for the other

Ru

Rd

Ru

Rd

LSR

Label Switched Path(LSP)

LSP defines the path through LSRs from ingress to egress router FEC is determined at the LER-ingress LSPs are unidirectional LSP might deviate from the IGP shortest path

LSP

LSP

Label

A short, fixed length identifier (32 bits) Sent with each packet Local between two routers Can have different labels if entering from different routers One label for one FEC Decided by the downstream router
LSR binds a label to an FEC It then informs the upstream LSR of the binding

Label (contd)

ATM
VCI/VPI field of ATM header

Frame Relay
DLCI field of FR header

PPP/LAN
shim header inserted between layer 2 and layer 3

Label (contd)

PPP Header
Layer 3 Header Label PPP Header

LAN MAC Header

Layer 3 Header Label MAC Header

ATM Cell Header

DATA HEC CLP PTI VCI VPI GFC

Label

Shim Header

TTL
31

S
23

EXP
22 19

Label
0

Label = 20 bits EXP = Experimental bits, 3 bits S = Bottom of stack, 1 bit TTL = Time To Live, 8 bits

Shim Header (contd)

EXP field
Also known as Class of Service (CoS) bits Used for experimentation to indicate packets treatment Queuing as well as scheduling Different packets can receive different treatment depending on the CoS value

S bit
Supports hierarchical label stack 1 if the label is the bottom most label in the label stack 0 for all other labels

Time To Live (TTL)

TTL value decremented by 1 when it passes through an LSR If TTL value = 0 before the destination, discard the packet Avoids loops may exist because of some misconfigurations Multicast scoping limit the scope of a packet Supporting the traceroute command

TTL (contd)

Shim header
Has an explicit TTL field Initially loaded from the IP header TTL field At the egress LER, value of TTL is copied into the TTL field of the IP header

Data link layer header (e.g VPI/VCI)

No explicit TTL field Ingress LER estimates the LSP length Decrements the TTL count by the LSP length If initial count of TTL less than the LSP length, discard the packet

Label stack

MPLS supports hierarchy A packet can carry a number of labels Each LSR processes the topmost label
Irrespective of the level of hierarchy

If traffic crosses several networks, it can be tunneled across them Use stacked labels Advantage reduces the LIB table of each router drastically

Label stack (contd)

Layer 2 Header

Label 3

Label 2

Label 1

IP Packet

MPLS Domain 1

MPLS Domain 2

MPLS Domain 3

Labels scope and uniqueness

Labels are local between two LSRs Rd might give label L1 for FEC F and distribute it to Ru1 At the same time, it might give a label L2 to FEC F and distribute it to Ru2 L1 might not necessarily be equal to L2 Can there be a same label for different FECs?
Generally, NO BUT no such specification LSR must have different label spaces to accommodate both SHIM header specifies that different label spaces used for unicast packets and multicast packets

Invalid labels

What should be done if an LSR receives an invalid label? Should it be forwarded as an unlabeled IP packet? Should it be discarded? MUST be discarded! Forwarding it can cause a loop Same treatment if there is no valid outgoing label

Route selection

Refers to the method of selecting an LSP for a particular FEC Done by LDP
Set of procedures and messages Messages exchanged between LSRs to establish an LSP LSRs associate an FEC with each LSP created

Two types of LDP

Hop by hop routing Explicit routing

Route selection (contd)

Hop by Hop
Allows each LSR to individually choose the next hop This is the usual mode today in existing IP networks No overhead processing as compared to IP

Explicit routing
A single router, generally the ingress LER,specifies several or all of the LSRs in the LSP Provides functionality for traffic engineering and QoS
o Several: loosely explicitly routed o All: strictly explicitly routed

E.g. CR-LDP, TE-RSVP

Label Information Base (LIB)

Table maintained by the LSRs Contents of the table

Incoming label Outgoing label Outgoing path Address prefix

Label Information Base (LIB)

Incoming label

Address Prefix

Outgoing Path

Outgoing label

MPLS forwarding

Existing routing protocols establish routes LDP establishes label to route mappings LDP creates LIB entries for each LSR Ingress LER receives packet,adds a label LSRs forward labeled packets using label swapping Egress LER removes the label and delivers the packet

MPLS forwarding (contd)

Address Prefix
Addres s Prefix Out Path In Label Out Label Address Prefix Out In Path Label Out Label

Out In Path Label

Out Label

125.5 0

125.5 0 145.40

125.5 0 145.40

8
125.5 2 0

0
3

Use label 9 for 125.50

Use label 2 for 125.50 and label 1 for 145.40
1 145.40

Use label 8 for 145.40

MPLS forwarding (contd)

Address Prefix
Addres s Prefix Out Path In Label Out Label Address Prefix Out In Path Label Out Label

Out In Path Label

Out Label

125.5 0

125.5 0 145.40

125.5 0 145.40

8
125.5 20

0
3

Data
Data 125.50.33.85 2
1

125.50.33.85
145.40

Multiprotocol Label Switching

Traditional IP Routing Disadvantages Need for MPLS MPLS basics and terminologies Experiments

Measurements of MPLS Traffic Engineering and QoS

Series of tests were run to evaluate the performance of TCP and UDP flows.
Tests include the effects of using different MPLS features on the performance of traffic flows.

Goals:
Evaluating how well MPLS traffic engineering and QoS can improve the performance of todays Internet. Identify opportunities for improvement and development of new mechanisms to ensure provision of traffic engineering as well as QoS/CoS features in future networks.

Experimental Network Configuration

Network Description

Host Computers:
Intel Pentium II, 300MHz processors, 128 MB RAM. Equipped with Fast Ethernet NICs and running FreeBSD 4.1. Connected to the MPLS domain using 100Base-T connections via Gigabit Ethernet switches.

Label Switched Routers:

Juniper Networks M40 routers running JUNOS Internet Software supporting Juniper Networks MPLS implementation. Routers connected using OC-12 ATM links. Distance between LSR1 and LSR3, LSR2 and LSR3 is about 40Km while LSR1 and LSR2 are 5Km apart.

Experiment Using MPLS Explicit LSPs

Minimize the effects of network congestion by using MPLS traffic engineering capability.
This is done by applying explicit routing.

Scenario 1:
Two explicit LSPs are established between LSR1 and LSR3, both following the IGP shortest path.

Scenario 2:
Two explicit LSPs set up again. However, traffic from host A to host C is made to traverse LSP2 while traffic from host B to host D flows across LSP1.

Results

Traffic from host A to host C is diverted to flow on the MPLS explicit path.

Significant improvement of throughput over the IGP shortest path is observed.

Throughput of TCP flow from Host A to Host C

Results (contd)

Throughput of both flows

Results(contd)

Average RTT is measured using Netperf request/response method.

RTT dramatically increases for congested IGP path, while it is minimal for packets traversing the MPLS explicit LSPs.

TCP average RTT

Results(contd)

UDP average RTT

Experiment Using MPLS CoS/QoS

Study how MPLS can be used to provide guaranteed bandwidth and different levels of service for flows.
This is done by characterizing each LSP with a certain reserved bandwidth across the MPLS network. Each LSP is also characterized with different CoS values.

Network configuration is set up in such a way as to apply MPLS service differentiation along the same path. Reservation of bandwidth is done using the Committed Data Rate (CDR) QoS parameter in CR-LDP.

Assigning CoS Values

EXP header is used. So, we have 8 different classes (0-7) to assign. A class indicates:
Output transmission queue to use, percent of the queue buffer to use, percent of link bandwidth to serve, packet loss priority to apply in presence of congestion. Traffic with higher priority class receives better treatment than a lower priority class.

Ingress router LSR1 is configured so that it can classify and map flows into LSP1 and LSP2 based on their destination address. The two LSPs are also configured with different CoS values.

Network Configuration For CoS Test

70 % bandwidth reserved for LSP1 30 % bandwidth reserved for LSP2

Bandwidth Reservation Over LSPs

This demonstrates how

we can reserve resources in advance, as well as ensure guaranteed bandwidth.

Results

Traffic from LSP1 is offered a higher service level and delivered with lower latency. Service differentiation using MPLS CoS values has a significant impact on the performance of applications.

Conclusion

Providing QoS and traffic engineering capabilities in the Internet is very essential. For this purpose, the current Internet must be enhanced with new technologies such as MPLS. MPLS will play a key role in future service providers and carriers IP backbone networks. The use of MPLS in IP backbone networks will facilitate the development of new services such as real-time applications in the Internet.