[ieee 2008 second international conference on electrical engineering (icee) - lahore, pakistan...
TRANSCRIPT
Second International Conference on Electrical Engineering
25-26 March 2008
University of Engineering and Technology, Lahore (Pakistan)
978-1-4244-2293-7/08/$25.00 ©2008 IEEE.
MPLS Feasibility for General & Core IP Networks using Open-Source
Systems Syed Adeel Ali Shah ([email protected]) Laeeq Ahmed ([email protected])
Department of Computer Science and Information Technology
N-W.F.P University of Engineering and Technology, Peshawar, Pakistan.
Abstract Over the past few years Internet growth has
become rapid and is now considered as primary
source of information sharing and communication.
This has resulted in development of new
applications and communication systems that
demand more service quality, reliability and
efficiency. These applications not only include
traditional data but also new applications for
voice, video conferencing along with peer-to-peer
applications. Current IP protocol suit is although
the dominant networking technology, but despite
its dominance, it has been exposed with
weaknesses regarding ever-increasing demands by
Internet applications. This paper surveys the
deficiencies of traditional Traffic Engineering
(TE) techniques used to overcome the
shortcomings of IP. Multiprotocol Label Switching
(MPLS) is looked into as an ideal solution for
General and Core IP Networks. Moreover,
Multiprotocol Label Switching (MPLS) is
implemented in a network by using Open Source
Kernels and tools to check its feasibility for IP
Networks.
1. Introduction
Primarily Traffic Engineering (TE) is a term,
which refers to the techniques used to manage
traffic in a network. Now a day Traffic
Engineering is considered to be inevitable for
networks that demand more reliability and better
management of network resources. Limitations of
traditional IP Routing, discussed in section [2], can
only be dealt with different Traffic Engineering
techniques.
Figure-1 shows a simple representation of a
network scenario having TE execution. Here node
R1 can reach R2 via three different paths i.e. A, B
and C respectively. Each link has different
bandwidth and can suite different categories of
traffic. Real Time Traffic can be forced to follow a
more bandwidth rich path and same can be the
case with VOIP and E-mail services as shown in
Figure-1. In this way network resources are
utilized more efficiently and traffic receives better
service.
Well-known techniques used for Traffic
Engineering are ATM-Overlay [1] and MPLS-TE.
ATM-Overlay model has some major drawbacks,
which are discussed in section [3] of this paper.
MPLS on the other hand is considered to be the
most appropriate solution for Traffic Engineering
encompassing the features of IP and ATM.
Internet is continuously on the rise in size ever
since it came into existence, so much so that every
year it increases almost by 100% [2]. Internet in its
current state has become unwieldy and
unpredictable in terms of providing specific
service for demanding applications. These
applications can be anything from web browsers to
Peer-to-Peer (P2P) and web2 applications [3]. P2P
however, is the main application that is
contributing to rapid growth of Internet [3].
According to functionality Internet can be divided
into two main parts i.e. one as the Core Network
and another as an IP Network.
Figure-2 shows a very basic division of the
Internet into Core and IP networks. Core Network
consists of ATM Switches having Mesh topology.
This network is connection oriented in nature and
provides reliable and efficient service for traffic.
However, major concern is the IP Network, which
2
lacks the performance efficiency of Core Network.
This is basically due to the complex architecture of
IP Network. Main Protocol operating in this
network is IP and it is very well suited for a
network of this kind with its Best Effort Routing.
However, Implications of Best Effort IP routing
and its incapability in Traffic Engineering is the
main concern here.
2. IP Routing Limitations
To better understand the importance of MPLS
as new TE technique we need to have an
understanding of IP Routing Limitations
Service Guarantee
Best effort routing otherwise means that there
can be no service guarantee for any kind of traffic,
which is forwarded in an IP domain. Service
Guarantee might not be an important issue for
lightweight traffic, but for time critical traffic e.g.
video and voice, service guarantee is an important
concern. Any delay or loss of data means more
jitters resulting in bad quality of traffic. IP
currently provides two methods of Quality of
Service (QoS) namely, Integrated QoS [4] and
Differentiated QoS. [5].
Class of Service (CoS) & Congestion
Awareness
IP Routing in its current state is not capable of
discriminating different types of traffic. All the
traffic in IP Routing domain is treated in the same
way regardless of the needs for network resources.
The CoS support on the basis of traffic or user is
not possible in IP, which eventually results in
under utilized or otherwise congested routes.
MPLS overcomes this problem with Forward
Equivalence Class support. Also, the Inability of
IP to find congestion or available bandwidth on a
route contributes in casual forwarding decisions.
Possible solution provided by IP to overcome this
problem is termed as Load Balancing, but due to
its simple criteria of dividing the traffic, it suffers
from scalability problem.
2.3 Scalability
IP Routing has encountered the Scalability
issue due to its routing mechanisms. In traditional
IP Routing, Control Plane (route determination)
and Forwarding Plane (traffic forwarding) are
closely integrated. Any change in either of these
two can result into overall change in the Routing
model. On the other hand routing algorithms in use
today operate slowly due to escalating size of
routing tables in routers. A report [6] issued in
2006 shows detailed analysis of routing table
sizes, which is continuously on the rise.
Improvement in IP routing model means combined
improvement in Control and Forwarding Planes
resulting in increased complexity, which does not
seem feasible in the context of routing table sizes.
At present, IP Routing execution in large networks
is inversely proportional to speed and hence is not
scalable.
2.4 IP Route Recovery
Route failures in complex Internet architecture
are common and there is an increasing need for
fast route recovery mechanisms. When a link goes
down in IP network its recovery speed depends on
three factors. First factor is the time taken by a
router to detect a link failure. Second factor is the
distribution of this information across its adjacent
router and effectively to the whole network. Third
factor is the calculation of new routing tables by
all routers and finding alternate route for traffic.
There are specific messages that Routing Protocols
use to determine whether a link is active or
inactive. However, in case of route failure routing
protocols cannot find the location of failed route.
These factors play an important role in making IP
rerouting techniques slow in convergence. Clearly
there is need for fast reroute techniques [7].
3. ATM Overlay Model & Limitations
ATM Overlay [1] model was considered to be
the superlative option for traffic engineering prior
to MPLS-TE. It is based on connection-oriented
model, which is crucial for guaranteed delivery of
data in core network. Consider Figure-2 where
data comes from IP Network into the ATM Core
Network. IP packets cannot be sent directly on
ATM routes, so the ATM Switches encapsulate IP
packets into fixed length ATM Cells. In this way
IP layer remains transparent to all ATM switches.
ATM can provide QoS with respect to different
parameters namely bandwidth, constant/variable
and unspecified bit rates. ATM arguably has the
best QoS model at present, but despite these
features Overlay model has some hitches, which
leads to the introduction of new technologies in
Core Network.
3
3.1 Bandwidth Utilization
Overlay model has low bandwidth utilization
due to the overhead involved in an ATM Cell.
Moreover ATM Switch has to encapsulate the IP
packet into 53 Bytes Cell. Figure-3 depicts an
ATM cell with 5 Bytes of overhead referred to as
cell tax, which decreases the capacity of a cell to
9.5% [8]. ATM does not support variable size
Cells and IP packet must be accommodated in
remaining 48 Bytes.
3.2 IP & ATM Network Management
Poor Integration of IP and ATM signify that IP
packets need to be encapsulated in ATM cells
before routing them on core network. It means that
ATM Overlay model holds two separate networks
i.e. IP and ATM and there is a need for separate
network management schemes for each. Also
signaling protocol for both IP (OSPF etc) and
ATM (PNNI) are different and needs to function
in their own domains. Network management of a
diverse network like ATM Overlay is very obscure
and complex.
3.3 Scalability
Figure-4 is the modified version of the MPLS
execution topology used in section [6.2], and here
it is used to explain scalability issue related with
ATM Overlay Model.
Figure-7 represents the physical topology of
Overlay model whereas Figure-4 shows logical
topology of the same model. ATM Overlay forms
a mesh topology where every node has a
connection to every other node.
Increase in the number of nodes means increase
in number of adjacencies making the network
architecture complex. More adjacencies in a core
network cause another problem, which is referred
to Flooding. Flooding occurs when there is a route
failure and every node has to forward state
information to every other node. As the network
grows in size Flooding becomes a major concern
as it floods the network with route failure
information.
4. Why MPLS for TE
MPLS is seen as a hybrid solution, which
encompasses the features of IP and ATM. The
most telling feature of MPLS is its ability to
separate Control Plane and Forwarding Plane [9].
This effectively means that TE is entirely
controlled by IP without any support of layer 2
technologies, which contributes to its simplicity.
Some of the features of MPLS-TE are discussed
here. MPLS plays an important role in the
introduction of Next Generation Network (NGN)
[10].
Temporary LSP Tunnels
MPLS as a hybrid solution inherits some
features from ATM and IP. One such feature that
correlates with ATM VCs (virtual circuits) is the
tunnel creation before communication between
two parties. Unlike IP with no service guarantee
for real time data, MPLS can provide some level
of guarantee by making sure that the Tunnel is
used by one service at a time. These tunnels are
not permanent as in ATM network and are
detached after communication. Also, QoS
achieved by ATM VCs can be provided using LSP
Tunnels.
Intelligent Routing decisions
Unlike IP routing, which is based on the
shortest path and destination IP addresses,
techniques used by MPLS for routing are much
sensible. Configuration of LSR/LER can be static
or dynamic and during configuration traffic and
path constraints are taken into account. In IP
Routing there are signaling protocols (OSPF, IS-
IS), which distribute path information without
considering path and traffic constraints. In ATM a
protocol called PNNI is used for the same reason.
MPLS also needs an efficient signaling
mechanism to share label information among all
LSRs. Constraints taken into consideration are
bandwidth, latency and type of traffic. MPLS also
allows for the classification of traffic according to
service needs. This classification is helpful in
correct assignment of network resources to each
traffic class. MPLS supports different protocols for
4
distributing constraint information across all
LSRs.
Scalability
IP and ATM suffer from scalability problem,
MPLS implements divide and conquer rule to
achieve scalability in case of IP routing. This
means that finding routes and routing the traffic
are distinct tasks. To send traffic across an MPLS
domain, first of all routing tables, made by routing
protocols are consulted for path information. Then
the traffic is forwarded, which does not involve
routing tables or routing protocols, but an
independent use of Shim labels. It means that
routing protocols are only limited to route
determination and increase in the network size
does not increase the load on these protocols
because labels are responsible for forwarding.
Figure-4 shows the logical topology of ATM
core network. In contrast to ATM network,
number of interconnections remains the same in
MPLS as the network grows in size. Figure-7
would be a more appropriate way of representing
physical and logical topology of MPLS domain.
MPLS addresses enhance the shortcoming of Load
Balancing in IP networks by introducing a concept
of Forward Equivalence Class (FEC), which
divides different types of traffic according to their
traffic requirements. Considering Figure-1, FEC
makes it possible to put all voice traffic into one
class e.g. A with low latency and more bandwidth.
And all mail traffic into class B with loose
network requirements. In this way MPLS can
force all voice traffic to follow route A and other
traffic to follow route B. FEC support for traffic in
MPLS represents better scalability than Load
Balancing.
Fast Reroute
MPLS implementation of route recovery is
much more proficient than IP. MPLS introduces a
concept of Label Stack, in which multiple labels
are applied in one shim label. Figure-5 is the
representation of a packet with multiple labels.
In MPLS there is a backup LSP also referred to
“Protection LSP”. Primary LSP serves as the
default route for Traffic from LER1 to LER2,
while Protection LSP acts as a redundant link for
default route in case of Route failure. When the
Primary LSP goes down all the traffic arriving at
LSR1 is examined for Label Stack. Label Stack
functions as an array with Last in First-Out (LIFO)
rule. Fast Reroute does not involve broadcasting of
failed route. This means that routers in an MPLS
domain are not entitled to recalculate new routes.
Instead new route information is carried in each
packet in the form of labels.
5. MPLS Implementation Resources
For implementation IBM P-4 workstations are
used having specialized Linux Kernel for MPLS.
Each system has two NIC cards for connectivity;
also, each system has MGEN, Netperf and
Ethereal installed for generating and analyzing
traffic patterns.
5.1 Network Design
Figure-6 shows the network design for
execution of MPLS. The topology forms an MPLS
domain with LER1 and LER2 as the core network.
Linux Fedora Core4 is installed on all nodes.
Moreover, specialized kernel for MPLS is installed
on LER1 and LER2. MPLS Kernel used here is
still under development for inclusion of advanced
features. It is an open source software project and
readily available for use and research purposes
[11].
Essential configuration for MPLS is done on
LER1 and LER2, which remains transparent to
HostA and HostB. IP addresses are assigned in
such a way that Figure-6 breaks up into three
different networks. Main idea is to send traffic
from either host across the routers to the other end
of the network. LER1 and LER2 behave as Ingress
and Egress routers depending on the direction of
traffic flow.
5.2 Experiment
Figure-7 is the network topology for this
experiment, which is basically similar to Figure-4
used for representing ATM Logical Topology.
5
Figure-7a is a network with static IP Routing
and Figure-7b shows the same network with
MPLS implementation. Idea of this experiment is
to use MGEN to load this network with traffic and
discover the Interarrival time for both networks
under same traffic load.
5.3 Execution
First, the Experiment is performed on a
network with static routing, Figure-7a. MGEN
resides on R1 and R3 and is used to generate UDP
traffic. Traffic pattern used by MGEN is UDP,
which sends 5000 Bytes of data every second for
duration of 90 seconds. Traffic generated from
LER1 is in two flows, the first flow initiates as
soon as MGEN runs while the second flow starts
after 30 seconds. MGEN uses Script files to
generate traffic on the source. R3 also requires an
instance of MGEN to capture UDP traffic and
store the relevant information in a log file. Trpr
can be used to parse the log file, generated by
MGEN, in order to find Packet Loss. Table-1a and
Table-1b shows the output generated after the
experiment.
5.4 Results
Table-1a shows the statistics for Interarrival
time and Table-1b shows Packet Loss for double
flow of data for the duration of 90 seconds. This
experiment is performed by static IP route
configuration in Figure-7a.
Same experiment is repeated in a network
shown in Figure-7b but with MPLS execution. The
implementation detail of MPLS remains the same.
MGEN is executed on LER1 and the results
captured on LER2 as shown in Table-2.
This experiment is repeated several times. In
Table-2a and Table-1a there is a small but notable
difference in values, although the amount of traffic
sent and time taken is the same in both cases. This
is an interesting result in Experiment, which shows
the times of two networks loaded with same
traffic. Although this report discusses the benefits
of MPLS in terms of its speed, but the results
shown in this experiment are opposite to what
MPLS can offer. There is a small but notable
difference between the Interarrival time of traffic
in both networks and MPLS time is on the higher
side. In view of the author this is because of the
32-bit overhead involved with every packet.
MPLS operates fast because it does switching,
however, label switching in this experiment is of
no importance. The reason being, it involves extra
processing in both LERs in order to check layer3
and then attach a label only to perform switching
in one LSR. Contrastingly results achieved with
static routes in same network are better because
the routing entries are not overtly populated. And
the routes only need to look up the next hop entry
without any need of attaching an extra label. With
this observation in functionality it is safe to say
that MPLS provides switching mechanism with
the cost of overhead. Therefore, MPLS is not an
option for small networks and for networks with
no Traffic Engineering requirements.
6. Conclusion
MPLS is an emerging technology and by no
means a perfect solution to current IP network
problems. It has got advantages and disadvantages
but at the moment it provides much better Traffic
Engineering capability than ATM Overlay. MPLS
is neither a substitute for IP Routing and nor it is a
new QoS model for existing carrier networks. In
actual fact MPLS operates in coordination with IP
Routing and its main objective is to provide the
speed of switching to Layer 3. Introduction of
Labels provide an effective alternative and evades
the need of large routing table lookups and results
in fast routing. However, the telling factor of
MPLS is its ability to manage and classify the
traffic in order to provide better utilization of
resources than IP Routing. MPLS is not designed
to compete with IP or ATM but rather to provide
better integration with the network layer and link
layer technologies. On the other hand both IP and
ATM have got integration problems. ATM mainly
serves the core network and it also needs to carry
the IP traffic in ATM cells. As MPLS supports
both technologies and hence can be used to
6
effectively resolve integration and traffic
engineering issues in carrier networks.
References
[1] Chuck Semeria "Traffic Engineering for the New
Public Networks",Juniper Networks, January 1999
[2] Andrew M.Odlyzko, “Internet Traffic Growth:
Source and Implications”, University of Minnesota,
Minneapolis, MN, USA, 2003
[3] http ://www.oreillynet.com/pub/a/orielly/tim/news
/2005/09/30/what-is-web-20.html
[4] B. Braden, D. Clark, and S. Shenker, "RFC1633:
Integrated Services in the Internet Architecture: an
Overview"
[5] S.Blake, D.Black, M.Carlson, E.Davies, Z.Wang,
W.Weiss, "An Architecture for Differentiated
Services"
[6] Philip Smith, "Internet Routing Table Analysis
Update", SANOG, Karachi, July 2006
[7] V.Sharma, F.Hellstrand, "Framework for Multi-
Protocol Label Switching (MPLS)-based
Recovery", February 2003
[8] Document ID:10489, "Cisco - Measuring the
Utilization of ATM PVCs"
[9] Bruce Davie, Yahok Rekhter, MPLS Technology
and Applications: Morgan Kaufmann, 2000
[10] Keith Knightson, Noataka Morita, Thomas Towle,
NGN Architecture: Generic Principles Functional
Architecture and implementation 2005
[11] http://sourceforge.net/