Beware the cult of IP
Network World:
The IP craze has hit an all-time high. The
slogan "everything over IP and IP over
everything" captures the zeal with which
many in the industry are pursuing the
expansion of IP-based features and services.
Everything that can possibly be crammed
into IP - or on top of it - is.
The next step is to go completely minimalist
- IP directly on fiber, or rather, over
photons. Who needs Layer 2? Just slap Layer
3 directly on the media. That's the vision of
some in the industry, including a Nortel
Networks executive who waxed philosophical
at the recent Internet Bandwidth (iBand)
conference about unleashing gigabits of
bandwidth in the public network. In his
vision, high-speed routers would blast
Ethernet frames over optical interfaces,
directly onto SONET rings or the fiber itself.
I understand the rush to embrace IP. All
projections show that data will be the vast
majority of traffic crossing networks over the
next five years. So if you're going to build a
new integrated services network, you would
optimize it for data.
What I don't understand is the fervor with
which these IP-based efforts are being
pursued. Many of the new features being
added to IP, such as class of service (CoS)
support and traffic management, already
exist at other layers in the network. For
reasons that are not strictly technical, these
existing technologies are deemed inadequate
and must be re-created in the image and
likeness of IP.
And so it will happen. A swarm of engineers
is working on the technologies for building a
multiservice Internet. IP-based CoS and
traffic management and IP-based telephony
will become realities. The issue is when - and
at what price to enterprise users.
As veterans of the Internet Engineering Task
Force point out, the IETF is driven from the
bottom up, not by a grand design. As a
result, the standards-making body produces
technology pieces, not integrated solutions.
The building of end-to-end technologies falls
to vendors, service providers and enterprise
network operators.
Right now, key IP multiservice technologies
exist only as evolving piece parts. Consider
Differentiated Services (Diff-Serv), the
newest IETF approach to CoS. Designed for
use in the Internet, Diff-Serv is elegant in its
simplicity. Each packet is marked with a
special flag (a series of bits) that indicates
how it should be treated. At each router hop
along the packet's path, the router sorts
packets into queues based on the flag. The
queues themselves get different treatment,
such as differing shares of bandwidth,
forwarding priority and probability of dropping
a packet in case of congestion.
To date, the IETF has defined few actual
classes of service based on Diff-Serv. Also,
there is room for interpretation of what has
been defined. Consequently, enterprise
customers and ISPs will need to write
service-level agreements (SLA) that clearly
spell out which applications get which types
of flags and how the ISP will actually act
upon those flags.
Likewise, if the packet crosses the networks
of multiple ISPs, these ISPs will have to
define compatible SLAs and handle the
packets in a comparable way for customers
to receive the end-to-end service for which
they've contracted. If two ISPs have
incompatible SLAs or support different
interpretations of the flags, the customer's
end-to-end service will be "squishy," in the
words of one of the Diff-Serv working group
co-chairs.
I don't know about you, but I don't think
squishy CoS is going to cut it for voice. Or
video. Fortunately, the sponsors of iBand,
Stardust Forums, had the foresight to hold a
meeting at the conference to discuss the
need for a QoS forum. Representatives from
nearly 50 companies attended the meeting.
There was general agreement that such a
forum is needed to clarify the business
drivers behind QoS, bring the relevant
technologies into focus and push
interoperability testing.
Such a forum is needed if IP-based CoS is to
become a reality on the Internet. If the
industry must rely on bilateral SLAs to
achieve end-to-end service, we'll never get
out of the squishy phase.
Bear in mind that CoS is only one new area
that the IETF is addressing. The organization
now has roughly 200 working groups pursuing
routing, addressing, security, policy and
other technical areas. My concern is that, in
this mad rush to embrace all things IP, we
may fail to appreciate - and exploit - the
capabilities of the installed base of
technology we already have.
IP is one technology tool among others and
can be used in a complementary fashion with
other technologies. It's not a savior.
Enterprise customers need to avoid getting
carried away by this wave of hype and focus
on the problems they need to solve today
and the spectrum of tools available to them.
Petrosky is an independent technology
analyst in San Mateo, Calif. She can be
reached at mary@mpetrosky.com.
[Copyright 1998, Network World]
|
