The Serve-to-Order IP Network Catering to a More Distinctive Palate
All-you-can-eat consumer
Internet service for $19.95
a month is alive and well.
But carriers are working to
provide businesses with
Internet protocol (IP)
services that are less
smorgasbord and more
serve-to-order.
"A huge number of
businesses understand that the broad public
connectivity of IP is literally going to change the
business world globally," says Rob Redford, product
marketing director at Cisco System Inc.'s
(www.cisco.com) multiservice switching business
unit. "Today's basic requirement of the Internet is
connectivity. You'd like security, QoS (quality of
service), but it's not required."
But as mission-critical business applications and
delay- sensitive traffic such as voice move to IP
networks, best-effort delivery no longer cuts it,
Redford says, so businesses are ordering carriers to
bring different service level options to the table.
"The question now is how do we evolve to that?" he
asks.
Of course, there's no single answer to that question.
But there are a variety of new technologies and
processes that promise to allow carriers to dish up
bandwidth, latency limitations and other performance
parameters to suit the palates of a variety of
customers and applications.
What's on the Menu
One of the key issues under discussion by the
Internet Engineering Task Force (IETF)
(www.ietf.org) is multiprotocol label switching
protocol (MPLS). According to most accounts,
MPLS is a derivation of flow switching and tag
switching, concepts presented to the industry by
Ipsilon (now owned by Nokia) and Cisco Systems
(www.cisco.com), respectively, a few years ago.
However, Ascend Communications Inc.
(www.ascend.com), Cisco and IBM Corp.
(www.ibm.com) authored the original IETF
framework document for MPLS, says Fred
Sammartino, director of IP product marketing for
Ascend.
MPLS attaches a label to a particular group of
packets that need to get to a particular destination.
The baseline idea of MPLS--for which the IETF is
expected to issue a first draft in mid-1999--is to
improve performance on IP router backbones
(which operate at Layer 3) by introducing Layer 2
switching. Layer 3 IP routing operates on a
software-based hop-by-hop basis, meaning it looks
up destination addresses for each and every packet.
Layer 2, meanwhile, applies to switching
technologies such as asynchronous transfer mode
(ATM) and frame relay, which rely on hardware to
quickly set a direct path for multiple packets headed
for the same place to reach their destination.
"MPLS is an admission that some sort of connection
orientation is needed in order to provide quality of
service [on IP networks]," says Steve Byars, chief
technical officer for Netrix Corp. (www.netrix.com).
Of course, the idea of expecting different carriers
along the total route of packets to implement MPLS
consistently, and/or having border protocol
adjustments to adjust for differences in carrier or
vendor implementation of MPLS labels is extremely
complex, says Dave Schriftgiesser, director of
marketing development for Lucent Technologies
Inc.'s (www.lucent.com) data networking systems
division.
"As a result, MPLS as an end-to-end value
proposition for the world's IP traffic is weak," he
says. "The other side of the coin is that, as more and
more backbone providers implement MPLS, traffic
flows do improve."
Although the IETF has not issued the final draft of
MPLS, several vendors already have come out with
products based on this concept. Ascend's IP
Navigator and Cisco's 7200 and 7500 routers and
LS1010 and BPX 8650 ATM switches (all of which
support tag switching) are examples of such
products.
Of course, MPLS is expected to offer more than
simply expedited routing. According to some
vendors, it also will let carriers engineer the core of
their networks. And carriers could potentially
leverage the MPLS label to offer a variety of
value-added services.
"MPLS in its most simple form does tag switching,
which speeds up the way the route lookup is done.
MPLS also does full mapping onto end-to-end
Layer 2 circuits. There's a big difference between the
two--a faster way to do routing vs. getting the
benefits of a switched core and still preserving the
Layer 3 IP interface," says Ascend's Sammartino.
Steve Onishi, product manager for the infrastructure
routing platform at Bay Networks
Inc.(www.baynetworks.com), also sees MPLS as a
way to engineer the core of networks. But he says
he doesn't see MPLS as an IP accelerator. Bay's
products are designed to support wire rates from the
beginning, he says.
Jeff White, vice president of marketing for routing
switch vendor Packet Engines Inc.
(www.packetengines.com), however, doesn't see
the value of tag switching at all.
"The whole reason for tag switching was because
routing was slow. As soon as wire speed routing hit
its stride, I saw interest in MPLS decreasing. MPLS
for us is hard to understand--does it really provide
more value?"
Yes, says Cisco's Redford. The most significant
benefit of MPLS is the ability to deliver value-added
services. Today carriers with private IP networks
ensure packet delivery by over-provisioning
bandwidth on those networks, which is the
expensive way to do it. These carriers want to
deliver new revenue-generating services, while at the
same time constraining their costs so they can reap
the profits of these new services.
MPLS and tag switching allow carriers to build
scaleable virtual private networks (VPNs) in a
connectionless IP network. And those labels or tags
can be set at the edge of the network to classify
certain transmissions as low latency or whatever.
"A good example of what can be done with our
extensions to MPLS is building connectionless IP
VPNs with multiple IP service classes over a
multiservice ATM network," he says.
Lucent's Schriftgiesser, however, notes that there's a
lot more to the story of improving packet loss
problems, delay problems, latency problems and
jitter problems in IP routing than MPLS.
"There are a lot of people working on these issues in
different, and possibly complimentary ways," he
says. "The new IP switches [that Lucent announced
in late May] can provide latency and jitter
improvements in a number of ways without
implementing MPLS. Some of these take advantage
of the switches' inherent speeds. Another approach,
MPOA (multi- protocol over ATM), has seen some
additional work that looks very promising for large
[public] networks."
An ATM Centerpiece?
ATM switching is not a requirement of MPLS and
isn't required to improve performance on IP
networks, but many carriers are using it because it
supports any kind of services from IP to frame relay,
says Redford.
"Companies like Level 3 [Communications Inc.] and
Qwest [Communications International Inc.] are
making a big deal out of IP, but all of them are
buying ATM switches," he says.
Newbridge Networks Inc. (www.newbridge.com) is
a strong believer in the power of ATM to drive those
value-added services.
According to Stu Aaron, assistant vice president of
marketing for the IP and internetworking group at
Newbridge, carriers need to offer business
customers more than just VPNs as they exist today,
which is basically as raw bandwidth. Businesses
need the ability to connect multiple sites without
having to invest in routers at all those sites, to
connect all their sites seamlessly to a homogenous
network, and to set those connections with security
and policies based on their needs.
"The nirvana of VPN is a true virtual routing service
with all your customizable policies," he says.
With ATM at the core carriers can deliver QoS,
Aaron says. "Mapping IP onto ATM lets you deliver
VPN with explicit QoS-- not just high and low
priorities. ATM defines 16 descriptions of quality;
ATM Forum in 4.0 spec defines them--latency,
delay variation and others. Those combine to be
QoS. Class of service is just a priority."
And ATM has well-defined policy enforcement and
call admission control, and makes it easier for the
service provider to deliver on service level
agreements (SLAs), he says.
Newbridge and its partners Siemens
(www.siemens.com) and 3Com Corp.
(www.3com.com) are backing an architecture they
call Carrier Scale Internetworking (CSI), which is a
standard framework for virtual private routing
services. Additional vendors will announce their
support for CSI at the Networld+Interop show later
this month in Atlanta, Aaron says.
The genesis for CSI, which currently is in trials with
undisclosed carriers, is MPOA, Aaron says. CSI
uses an ATM core and tools to build virtual private
services around it. It uses ATM as the backbone
architecture and puts whatever applications or
protocols at the edges, Aaron says.
"MPLS is a protocol for connecting more routers
more efficiently. Within CSI, protocols like MPLS
are used. CSI is to MPLS as Federal Express is to
ZIP code," he says. "Fed Ex is a service, but
requires a ZIP code to deliver that package most
efficiently."
Differentiated Services, or DiffServ, is another set of
standards closer to CSI in concept, he says.
"We're looking at ways to envelope DiffServ in
CSI," he adds. "DiffServ is trying to address similar
problems as CSI but on a smaller scale."
The VIP Table
DiffServ is another standards effort within the IETF.
The idea behind DiffServ is to define the existing
type of service (ToS) bytes in the IP packet header
so the network knows the priority of the traffic.
According to Byars of Netrix, the beautiful thing
about DiffServ is it can be backward- compatible
with IPv4 (the version of IP that's generally
implemented).
But, on the down side, settling on a definition for
those bits has been a significant challenge,
Sammartino says.
"The concept is simple, but I've seen a huge swing in
what each of the bits mean," Sammartino says. Most
recently, there was discussion of mapping six bits
into a matrix with a variety of QoS parameters, he
says, but no application could possibly pick a
priority based on this matrix.
Another issue is that even if the network can
distinguish high-priority packets from low-priority
packets, it still doesn't provide QoS guarantees, he
says.
"It's just like flying on an airplane--you [can] get the
better seat or the better meal, but you're still not sure
the plane will take off," he says.
If there are a lot of high-priority packets on the
network--and that's likely to happen considering
most people will want to mark their transmissions
high priority--packets still can get bumped, he says.
"DiffServ at the edge sorts traffic into different
service classes," says Onishi of Bay, which is
implementing a pre-standards version of DiffServ in
its Versalar 15000 edge router. "You want to
rate-control premium traffic entering network at the
edges. In the backbone you look at the ToS field
and use that in the backbone routers to dump it into
queues so [packets are] served in priority order.
Does priority mean strict priority or weighted fair
queuing? I'm not sure if DiffServ defines a queuing
mechanism."
Other key challenges to implementing DiffServ will
include updating applications to use the bits, and
policing the network to make sure it is acting on the
ToS bits as required, Sammartino says.
"It's pretty complicated," he says.
Give Me the Usual
Carriers and their vendors
envision a day when
servers holding policies on
particular users or
applications will sit on both
public and private
networks to provide policy
parameters to the routers
and switches as needed.
This concept is widely
know as the
directory-enabled network (DEN).
Going forward, as DEN comes into strong play,
routing switches will act as LDAP clients and could
be managed by a DEN group, says Kevin Sheehan,
director of product management for Packet Engines
(www.packetengines.com). DEN, which is being
driven by Cisco and Microsoft Corp.
(www.microsoft.com), will come into the network
next year, he says.
"DEN takes policies and maps them to networking
switching," says Aaron of Newbridge. "In the old
days the way you dealt with getting traffic from Point
A to Point B is you looked at the Mac layer (Layer
2 in the OSI model) or IP addresses. Now you can
go higher than the IP address; now you can do that
based on user name, location or application."
That type of network will give enterprise network
managers more control in managing their networks.
Carriers, meanwhile, will be able to serve up--and
bill for--a variety of new services based on the
specific needs of customers and their applications. |
