8/98 BoardWatch Cover Story. Juniper Team Appears Ready To Challenge For Control Of The Internet Core
[What does everyone think about a LU/ASND/Juniper combo [Note the 10/5-10/6/98 WSJ NYC conference participants list]? Hasn't ASND also been very early on the MPLS bandwagon compared to the CSCO focus on tag switching?]
boardwatch.internet.com
by Bill McCarthy
Juniper Networks looks like a company that will give Cisco
Systems a run for the money in Internet routing. Juniper has lots of money -
money that reflects a faith in its team to come up with a better way to route
the core of the Internet.
And what does Juniper have faith in? MPLS, also known as multiprotocol
label swapping.
Three system and IP architects founded Juniper Networks in February 1996
in Mountain View, California, with a fervent belief in MPLS religion. Since
then Juniper has drawn managers and engineers from companies like Bay
Networks, Cisco Systems, Silicon Graphics, StrataCom, Sun Microsystems,
3Com, and Xerox. They brought in CEO Scott Kriens in September 1996,
and since then the momentum has been building in an equation that fits
Moore's Law.
THE MONEY
That momentum has drawn some of the largest players on the Net to the
MPLS faith, or at least to the point of making sure that they are seen in the
church - just in case. Look who else has the MPLS religion: AT&T
Ventures, Ericsson Inc., Lucent Technologies, Nortel (Northern Telecom),
the Siemens/Newbridge alliance, 3Com Corporation, The Anschutz Family
Investment Company LLC and WorldCom Inc.'s subsidiary, UUNET
Technologies, Inc. The big boys have poured $62 million into Juniper based
solely on its potential - although software and hardware products should be
making an appearance this summer. Juniper also formed a strategic
technology relationship with IBM.
If you look at Juniper's line up and the technology, while taking into account
where the Internet must go to deliver the promised services of tomorrow, you
may find yourself at the alter of MPLS as well.
The Anschutz Family Investment Company LLC, for example, invested $2.5
million in Juniper Networks. The Juniper investment joins other investments by
Anschutz in technology companies, including a majority stake in Qwest
Communications International Inc. Qwest is installing huge pipes for its
domestic network to connect 125 cities in the coming year. Qwest says that
its capacity represents about 80 percent of the data and voice traffic
originating in the United States plus
____________________________________________________________
(continued from front page)
planned network extensions 1,400 miles into Mexico. But huge
investments in fiber capacity can, of course, be wasted without
high-speed routers. "At Qwest, we're deploying OC-192 technology to
transport high capacity data traffic," said Joseph P. Nacchio, president and
CEO of Qwest, in a press release. "It's encouraging to see companies, like
Juniper, accelerating the development of router technology that can originate
and terminate this traffic at the highest possible speeds."
In addition to ownership interests, the participating companies have the
opportunity to integrate Juniper's technology with their existing product lines
and services worldwide. AT&T Ventures, for example, specifically invests in
information technology and service-enabling companies in emerging growth
markets, and helps companies that it believes will eventually help AT&T.
Obviously UUNET and Ericcsson, which is also making its own MPLS
routers, have obvious stakes in the future of the technology, as well.
THE TEAM
Juniper's founders all come from the industrial side of the Net; they're techies.
Chief Technical Officer Pradeep Sindhu, a principal scientist at Xerox PARC,
was a key architect of Sun's first high performance multiprocessor systems.
Sindhu's research at PARC focused on design tools for VLSI and high-speed
interconnects for shared-memory multiprocessors. That research led Sun
Microsystems and Xerox to develop Sun's first high-performance
multiprocessor system family, including the SS1000, SS2000, SS1000E, and
SS2000E. Sindhu, who has a Ph.D. in Computer Science from Carnegie
Mellon University, played key roles in the architecture, design, and
development of those machines.
The other two have impressive credentials as well.
Dennis Ferguson comes from MCI. While working for MCI and Advanced
Network Services, Ferguson worked in the development of routers and the
deployment of router technology in large Internet backbone networks. At
MCI, Ferguson implemented the vBNS research network and planned MCI's
commercial Internet service. Ferguson also developed significant
enhancements to the GateD suite of routing protocols at Advanced Networks
and Services, and developed routers for the original national backbone for
Canada - CA*net.
Bjorn Liencres was hardware technical lead at Sun Microsystems and an
architect of Sun's Ultra Enterprise family of servers. At Sun, Liencres also
served as Application Specific Integrated Circuits (ASIC) technical lead on
the SC2000 and the SS1000. Prior to Sun, Liencres worked for IBM, and he
has 10 patents filed or granted. (By the way, IBM is providing custom ASICs
for Juniper's new class of Internet backbone devices as part of a strategic
technology alliance. Under the agreement, IBM is custom designing the chips
for Juniper's products, integrating Juniper's software and IBM silicon logic).
__________________________________________________________
(continued from front page)
The three founders have surrounded themselves with some of the
top managers available. Chairman, President and CEO Scott
Kriens was a founder of StrataCom and spent 10 years as a vice president of
sales and operations, bringing in the first sales of that company. [CSCO brain drain...] Some give
Kriens credit for establishing the first Frame Relay and ATM networks in the
industry. Kriens also worked for Tandem Computers.
Chief Financial Officer Marcel Gani served as vice president and chief
financial officer of NVIDIA Corporation, Grand Junction Networks, Primary
Access and NeXT Computer, as well as doing 12 years at Intel Corporation.
Vice President of Engineering Peter Wexler jumped from Bay Networks
where he was a vice president of engineering. Both Joe Furgerson, director of
marketing at Juniper, and Gary Heidenreich, vice president of operations, are
from 3Com. Steven Haley is vice president of sales and jumped to Juniper
from Cisco and StrataCom. [COMS/BAY brain drain...]
THE TECHNOLOGY
High-performance routers are sometimes called Layer 3 switches, and the first
generation of Layer 3 switches used proprietary technology, so the Internet
Engineering Task Force formed the MPLS working group to formulate the
Multiprotocol Label Switching standard. That will allow the devices to
interoperate at some point, and the MPLS standard may be issued sometime
in 1999. In the technology, labels are associated with specific streams of data
and forwarding is simplified by the use of short fixed length labels to identify
streams rather than each packet.
In its simplest form, MPLS is an attempt to blend the best of the IP protocol
with the cell switching technology used in the ATM protocol. In fact, early
implementations created virtual circuits on ATM networks for sessions of
TCP/IP packets. The machines calculated only the route for the first packet in
a given transmission, sending the rest of the packets along the same path.
MPLS retains the concept of routing only the first packet in a stream or
session, but it can do away with the need to create numerous virtual circuits as
it does within ATM implementations. ATM's small cell size of 48 bytes also
eats unnecessary bandwidth by increasing the header to payload ratio. So it is
inefficient for many applications. Instead, MPLS attaches labels containing
condensed forwarding information to packets, so routers will know the next
hop.
With MPLS, only the edge routers that are connected to other ISPs'
networks need full routing tables and only the routers on the edge of the
network need to calculate routes. Currently, the Internet relies on each IP
datagram being routed separately with complex processing at each hop. The
IP routing architecture sees a network as a collection of routing domains.
Within a domain, routing is provided through interior routing - OSPF
-while routing across domains is provided through exterior routing - BGP.
However, all routers within domains that carry traffic transiting a network have
to maintain information provided by both interior routing and exterior routing.
The amount of information is significant and uses resources and time.
___________________________________________________________________
(continued from front page)
Tag switching allows separation of interior and exterior routing. With tag
switching only tag switches at the border of a domain need routing information
provided by exterior routing - all other switches within the domain only need
information about the interior of the network. That reduces the load on interior
switches, and shortens routing convergence time. To support this functionality,
tag switching allows a packet to carry a set of tags within a stack, and that
information can be used in a variety of ways. Label swapping allows packet
forwarding to be based on a match for a short label for a stream of packets.
Internal routers and switches examine the MPLS label, which gives the
address of the next edge router the packet must travel to. By cutting down on
the number of route calculations, overall traffic capacities and speed is
supposed to increase. And labels may also be used to deliver quality of
service by setting priorities for packets, so that guaranteed levels of bandwidth
can be given as needed.
It is also possible to bind a tag not just to a single route, but to a group of
routes, creating many-to-one mapping between routes and tags. Tags can be
carried in a number of ways, including as a small header inserted between the
Layer 2 and the network layer headers; as part of the Layer 2 header, if the
Layer 2 header provides adequate semantics as in Frame Relay, or ATM; or
as part of the network layer header.
Tag switching consists of forwarding and control. The forwarding component
uses the tag information carried by packets and the tag forwarding information
maintained by a tag switch to perform packet forwarding. The control
component is responsible for maintaining correct tag forwarding information
among a group of inter-connected tag switches. Segregating control and
forwarding into separate components promotes modularity or greater
flexibility in making adjustments to accommodate new requirements.
It is possible to implement tag switching over virtually any media type including
point-to-point links, multi-access links, and ATM. The tag forwarding
component is network layer independent. Use of control components specific
to a particular network layer protocol enables the use of tag switching with
different network layer protocols. In some cases MPLS may make direct use
of underlying Layer 2 forwarding, such as is provided by ATM or Frame
Relay equipment. Labels may be distributed to allow nodes to determine
which labels to use for specific streams or may use some sort of control
exchange, be piggybacked on a routing protocol, or both.
A large number of vendors are involved in developing MPLS, including the
usual suspects: Cisco Systems, 3Com, Bay and Ascend Communications.
And we are bound to hear a significant debate about the IETF standard and
which product is faster, better, and so on. But Cisco is still the company to
beat with most ISPs and backbone providers using Cisco products at the
core of their networks. [Really?...] And it will be difficult to unseat the champion. If a
company is suppling its customers with a product that works based on
products that work from a vendor, why change?
But then again this is the Internet, where Moore's Law - the rule of 18
months to obsolescence - may even be behind the time curve, and
competition and cooperation at times seem indistinguishable, so a number of
the big players seem to need to hedge their bets. They look at Juniper
Network's team and the technology, and they have a little faith - $62 million
worth. |
