|
Archive
Flanagan Consulting
Network Analysts and Consultants
"We Have the Experience"
-------------------------------------------------------------------------
ViewsLetter on Provisioning
2 Dec 2002 #11
IN THIS ISSUE:
>> How MPLS fits into provisioning, Part 2: Encapsulation Methods
============ + ===== + ===== + ===== + ============
>>Analysis
PROVISIONING SERVICES WOULD BE SIMPLER
IF ONE NETWORK CARRIED EVERY PROTOCOL
Some people talk as if a universal network were only weeks away, at which
time simplification of provisioning will create nirvana. If only
everybody's traffic of every kind could be put on the same backbone,
carriers wouldn't have to maintain and track the configuration of Frame
Relay, ATM, ISDN, POTS, SMDS, Telex, TDM leased line, and SONET
networks--just IP. Does that mean carriers could discontinue all
but one
service? Not likely. So far they've been able to turn off
only one of
the services ever introduced in North America.
[WIN-A-BOOK CONTEST: First three readers who email the name of a
carrier
service that has been completely discontinued--that is, no longer operates
at all--will receive a copy of "T-1 Networking" by William A. Flanagan.
Winning entries must include postal address for delivery of book.
If you
don't win, order a copy from our web site, under Publications.]
The point is that even a universal network will offer different services.
How? By emulating other networks at the edge. The backbone
will have to
encapsulate legacy formats in "the one" transport protocol.
MultiProtocol Label Switching (MPLS) pops up in these discussions for
several reasons:
--MPLS came out of the IP standards groups; those affiliations lend
it
credibility with the younger networkers who will implement it.
--MPLS is connection-oriented, providing ways to ensure quality of service
that aren't available to basic IP datagrams.
--Configuration of MPLS connections can be automated by applying familiar
IP routing protocols (with extensions, plus additional protocols:
MPLS
doesn't run on pure-IP networks). Generalized MPLS (GMPLS) handles
TMC
channels and optical wavelengths, too.
--If routing protocols extend to customer premises equipment, the customer
essentially configures his own services, achieving a good portion of the
potential savings anticipated from convergence.
The last two points make MPLS a candidate for automated provisioning (the
reason we cover it in this ViewsLetter). What makes MPLS practical
is its
power to carry legacy data formats. "Old data" must be put into
"new
headers"--that's how MPLS provides a place to carry the label ("label
switching" was described last time).
Simplest MPLS encapsulation is the addition of a 4-byte "shim header"
between the Layer 2 header (Ethernet MAC, PPP) and an IP (Layer 3) header,
expanding a "standard" IP packet. The shim contains 20 bits for
the
label, 3 bits reserved for experimentation (currently Class of Service),
a
1-bit Stack (S) field to indicate a shim header is the last one before
the
IP header, and a Time To Live field (1 byte). The Label Edge Router
(LER)
inserts the shim into packets arriving from an IP network and headed for
the MPLS backbone of Label Switching Routers (LSRs). MPLS packets
lose
the label before handoff to an IP network.
When an MPLS packet is forwarded along the Label Switched Path (LSP) by
an
LxR, the LxR may either "label swap" or preserve the existing shim header
and put the packet into a tunnel LSP defined by an additional shim header
with its own label. The S bit on the original (single) header indicates
it is last in the stack.
Note that the "shim" and "label switching" take no precautions about frame
ordering. A higher-level protocol may provide that service.
However, Frame Relay and ATM explicitly promise to preserve the order
of
packets. To emulate these connection-oriented services, MPLS uses
a
4-byte "control word" containing a sequence number. There are spaces
for
semaphore bits and a length indicator. When encapsulating a Frame
Relay
frame, the control word replaces the FR header--it carries the FECN, BECN,
D, and C bits from the FR header.
On top of this control word the LER adds the label header or headers.
At
its outbound port, the LxR also adds a link-layer protocol header such
as
PPP, Ethernet MAC, etc.
Adding headers uses bandwidth, so economies are taken where possible.
The
FR header and CRC aren't encapsulated, only the payload. ATM cells
each
lose one byte (the header error check, replaced by a higher-level CRC).
Multiple cells may be packed into one MPLS frame, reducing the
packets-per-second processing burden on the LSRs. Further, cells
can be
reassembled into the Common Part Convergence Sublayer (an AAL frame plus
length indicator) to eliminate the "cell tax" of a cell header for each
48
bytes of payload.
Ironies exist in this scheme of link-layer encapsulations. Adjacent
LSRs
most often use PPP at the link layer for WAN connections, Ethernet
locally. Yet MPLS defines ways to encapsulate PPP and Ethernet frames
in
MPLS, which then go into PPP or Ethernet to reach the next LSR.
For IP
this is silly, but it does offer the flexibility for I/O ports to handle
IXP, Vines, SNA, and other protocols without knowing it.
By also using the experimental bits in the control word to indicate Class
of Service (COS, essentially priority), the mechanism described for FR
and
ATM can emulate a Time Division Multiplexed (TDM) circuit. A packet
connection with high priority, some buffering at the receive end to smooth
out jitter, and synchronized bit clocks at both ends (or clocks that
adjust as needed to avoid over/under running the jitter buffer) can be
hard to tell from a digital private line.
Things are coming together, but not next week. The first step has been
achieved: combining voice and data on one infrastructure.
This
achievement is far from a universal solution, however. In many cases
the
"multiprotocol" traffic is limited to IP and Voice over IP. Between
sites, such a network still may use frame relay or leased lines.
As a possible transition, the standards define MPLS over "native" FR and
ATM networks. The label is simply the DLCI or VCC--the same forwarding
but with "MPLS procedures" to distribute the labels to the switches (that
is, to set up the virtual circuits). Nobody seems much interested,
as the
bandwidth saved doesn't represent a significant cost amid a bandwidth
glut. The major costs are elsewhere.
The desire to reduce capital expense drives long-term network planning
toward "convergence." An even larger cost element, as carriers are
now
appreciating, is operating expense--something that automation in
provisioning addresses directly. Automation in MPLS involves signaling,
routing protocols, and traffic engineering. We'll start on those
topics
next time, when we'll see where routing tables come from and how they
relate to service provisioning.
"Flanagan Consulting" and "ViewsLetter" are
Service Marks of W. A. Flanagan, Inc.
Updated: 11 June 2003
|
