Cisco unveiled, pity the scars....
07:32am EDT 13-Aug-01 Dresdner Kleinwort Wasserstein (Mahler, Ariane)
CSCO: Unfulfilled promises
Cisco's traditional dominance of the IETF came under pressure during last
week's meetings in London. Several speakers acknowledged the flaws inherent in
today's version of IP, DiffServ and MPLS, and carriers reiterated their
concerns over security and reliability of Internet-based solutions. In an
environment where traffic engineering and returns on invested capital are key
considerations, carriers are more outspoken about questioning the value of
Cisco's approach. With incumbent service providers no longer threatened by
competition, it is on their own terms that new protocols will be adopted. We
continue to rate Cisco Reduce, with a $14 per share target price.
Summary
The Internet Engineering Task Force (IETF) meetings in London last week
provided some insights as to the carriers' new requirements for Quality of
Service (QoS) and their reasons for delaying adoption of IP-based
architectures. While certain Cisco speakers presented ideas for addressing
carriers' QoS concerns, overall these solutions were perceived as either
inadequate or little more than concepts. In addition, several carriers
complained about routers' inherent lack of scalability and reliability and
blamed IP for failing to provide adequate returns on their existing
infrastructure. As the need for new protocols was discussed, incumbent service
providers seemed further away from embracing Cisco's IP-based solutions. This
provided a fertile ground for other suppliers to present more suitable
approaches for traditional carriers based on traffic engineering (TE).
With Cisco's current stock price discounting 25% top line growth and 20%
operating margins in 2002 and beyond, we are hard-pressed to find the sources
of such exceptional growth, especially in an environment where incumbent
carriers are dictating a slower speed of adoption of new technologies. With the
Enterprise sector unlikely to sustain such revenue and margins, we find
Cisco's stock price to be significantly overvalued. We therefore continue to
rate Cisco Reduce, with a $14 per share target price.
Key themes
Far from dominating the IETF meetings, as had been the case historically, Cisco
found itself facing greater skepticism than ever before at last weeks'
sessions in London. Most of the discussions focused on the shortcomings of IP,
MPLS and DiffServ, and the need for new traffic engineering protocols. Traffic
engineering, defined as the optimization of traffic flows so as to maximize
network utilization and minimize or eliminate congestion, was highlighted as a
key area of current and future work. The inadequacy of existing Internet
protocols was viewed as particularly problematic in an environment where
carriers seek to preserve QoS while leveraging their existing infrastructure.
While this may not have been the case with Cisco's emerging carrier customers
last year, we believe that incumbent carriers are now in the driver's seat to
dictate the terms under which VoIP and MPLS can be adopted. We came away
feeling that more obstacles had been discussed than solutions, and that Cisco
needed to go back to the drawing board to develop an approach more suited to
the traditional service providers.
AT&T on IP and MPLS
In a paper presented by Jerry Ash of AT&T (T-$19.58), routers were blamed for
lack of scalability and inability to deal with failures. In his own words,
"there have been a number of major outages reported by most major carriers, and
routing protocols have generally been involved." AT&T argued that there was
evidence that current routing protocols such as OSPF and ISIS could not recover
from large failures that resulted in widespread loss of topology database
information. When routes were finally recomputed, they were based on incomplete
topology recovery and needed to be recomputed again frequently. AT&T cited
several examples in which massive failures had been experienced and manual
intervention was required to prevent further flooding and to start recovery
procedures. Much work was needed, in AT&T's opinion, to address congestion
control and failure recovery in IP-based networks. AT&T went on to present
potential solutions, but stressed that its proposal was only a high-level
discussion meant to initiate further debate.
On the subject of MPLS and VPNs, two prominent AT&T Labs researchers (Steve
Bellovin and Randy Bush) pointed out that MPLS would create serious network
management challenges, together with unprecedented issues of privacy and
security. Among other direct criticisms of Cisco's approach, these two
researchers argued that MPLS was unnecessary for carriers able to run ATM or
frame relay directly over an internet backbone, and that VPNs would neither
scale nor provide security.
Cisco on DiffServ/VoIP
The paper which caught most of the attention was presented by Francois le
Faucheur of Cisco, jointly with Level 3 Communications, Global Crossing,
CoreExpress and Tenor Networks. This paper explored the need to add traffic
engineering mechanisms to DiffServ. The author acknowledged several
difficulties with current VoIP and DiffServ protocols, and offered solutions
which suggested that further work would be required on concept definition,
standard acceptance, design of hardware and software. The problems highlighted
were as follows:
1. Voice: An IP/MPLS network may need to carry a significant amount of voice
traffic relative to the capacity of its various links. For example, 10,000
uncompressed calls at 20 ms result in approximately 1 Gbps of IP traffic,
which is substantial relative to a traditional OC-48 link capacity (2.5
Gbps). In order to minimize delay and jitter, it is undesirable to carry
more than a small percentage of voice traffic on any link. In order to
satisfy carriers' need for greater capacity utilization, new traffic
engineering mechanisms must be explored to give voice priority over other
classes of traffic subject to a maximum pre-defined link utilization ratio
(25% or less).
2. Failure: Under failure of some links, the remaining links may not be
sufficient to ensure that after rerouting, high priority traffic does not
exceed the acceptable percentage. For example, high priority traffic
representing 25% of an OC-48 link may be rerouted on an OC-3 or 12 link in
case of a failure but then exceed the capacity of such links. This would
result in unacceptable degradation of quality of the high priority traffic.
Cisco concluded that different bandwidth constraints would need to be
defined for each class of service, suggesting much greater work was required
before reliable systems could be offered.
3. Guaranteed Bandwidth Services: Given carriers' preoccupation about
generating returns on their investment in infrastructure, best-efforts
services usually associated with IP look destined to receive less interest
than guaranteed bandwidth services, whereby the carrier commits itself to
providing a specified level of performance with respect to bandwidth, delay,
packet loss etc. Unfortunately, most of these services are difficult to
offer beyond a point-to-point configuration. Non-guaranteed traffic needs to
be subject to less stringent constraints and guaranteed traffic needs to
remain below a certain level of overall capacity. Cisco acknowledged that
existing protocols could not address these requirements.
4. Scalability: With traffic engineering enforcing different constraints for
different classes of traffic, scalability issues emerge when different class
types must be supported. It was not clear whether overhead could be kept to
a minimum with so many class types and classes within a class type to be
supported.
5. Security: In its paper on how to complement DiffServ with TE, Cisco
acknowledged that security issues were not addressed. While DiffServ may
offer some protection against denial-of-service attacks, Cisco suggested
that it was conceivable for lower priority traffic to be more susceptible to
security problems.
Lucent, WorldCom and AT&T on MPLS
Lucent, WorldCom and AT&T delivered a paper written to influence the direction
of MPLS toward greater standard compatibility between survivability approaches.
A design team presented ideas on the need to define a hierarchy between metro,
access, long-haul before adopting MPLS. The issue of survivability was
addressed with a self-healing approach combining spare capacity and diversity
in dealing with network failures occurring at different levels of the network
hierarchy.
Scalability issues of MPLS were also addressed. Simply put, to achieve full
protection and restoration, the number of label-switched paths required is N2
where N is the number of nodes.
A distinction was established between "vertical hierarchy," defined as the
communication between network layers such as TDM, optical and MPLS, and
"horizontal hierarchy," defined as between two areas or administrative
subdivisions within the same network layer. The design team concluded that the
most pressing need was with respect to horizontal hierarchy within the context
of Layer 2 and Layer 3 VPN services. In this scenario, service level agreements
necessitate signalling from the edges into the core of the network. The
limitations of current protocols such as multi-area OSPF were pointed out.
The interesting aspect of this paper was the attempt to highlight trade-offs
between different approaches as opposed to promising solutions based on
proprietary protocols yet to be defined and agreed on. The latter approach,
favored by Cisco since the early days of its involvement with IETF, no longer
seems to be acceptable to carriers, the majority of which are now of the
incumbent category. For example, the design team pointed out that protection
techniques, based on having a dedicated entity set up prior to the failure,
offered fast recovery from failure, while restoration techniques, which rely on
the use of spare capacity in the network, usually achieve better network
utilization. The recommended approach was to consider a spectrum of protection
and survivability mechanisms, which allow an operator some flexibility within
the context of a multi-vendor network. As to Cisco's IP + optical approach, the
design team felt that the benefits of merging the control planes of multiple
layers were far from evident at this stage.
Conclusion
With Cisco's current stock price discounting 25% top line growth and 20%
operating margins in 2002 and beyond, we believe that little attention has been
paid to Cisco's role in an environment where incumbent carriers rule on
adoption of new technologies. As the above analysis shows, Cisco is no longer
in a position to dictate the terms of new protocols when it is targeting
carriers with fundamental concerns for traffic engineering, reliability,
scalability and security, areas in which Cisco is viewed as having
over-promised and under-delivered. With the Enterprise sector unlikely to
sustain such revenue and margins, we find Cisco's stock price to be
significantly overvalued. We continue to believe that Cisco's intrinsic value
is $14 per share, which is derived using a 40 times multiple on our $0.35
estimate for fiscal 2003. We therefore reiterate our Reduce rating on Cisco
Systems. |
