|
|
Archive
----------------------------------------------------------------------------------
ViewsLetter on
Provisioning 28
Sept
2004
#41
----------------------------------------------------------------------------------
Provisioning automation--from chips to the business layer.
RESILIENT PACKET RING: LIKE ETHERNET,
BUT WITH SONET RECOVERY TIME (50 MS)
By William Flanagan, Editor & Publisher
Ethernet, the hands-down winner for Local Area Networks, appeals for
Metro and Wide Area connections as well. Available bandwidth is
high enough to overcome the inefficiency of using so many
headers--Media Access Control as well as IP, TCP, and probably
others--to carry even one byte of payload. Simplicity and
compatibility trump efficiency.
But wait! What if you could have it all? Not
satisfied? Want more for your money? What if I could add
the speedy recovery of a SONET ring when a cable is cut? Tell 'ya
what I'm gonna do!
No, not finish a hokey carnival pitch. We will look at Resilient
Packet Rings as a technology for the local loop. No rocket
science, so we'll stick with text. For pictures, refer to
www.rpralliance.com and the sites of members listed there.
FAMILIAR BASICS
If you're reading this, you should know at least the basics of
Synchronous Optical NETwork (SONET), a workhorse of local access and
metro area networks (MANs). The key concepts are:
--ring topology: traffic between any two points on the ring may
travel in either direction around the ring.
--add/drop multiplexing in the nodes: any point on the ring can
be the end of a connection, and bandwidth on the ring is assigned
independently in each direction.
--self-healing: when nodes on the ring detect a break (cable or
node), they send traffic in the opposite direction to avoid the fault.
When T-1 multiplexers were new (the 1980's) the terms used were "drop
and insert" and "alternate routing" but the effect was similar--Time
Division Multiplexed (TDM) channels restored automatically. Now
we have a way to the job with almost-Ethernet packets: RPR.
Again the terminology changes, but the functions are similar.
RPRs have dual fibers, one for each direction. Nodes, up to 255
per ring, sit on both fibers and act as 2-port, bidirectional,
store-and-forward switches (or routers).
By design, the RPR nodes receive and re-inject each packet passing
through. This gives each node the power to buffer a
lower-priority packet as it arrives from another node, so a local
packet of higher priority can be sent first. This ability prevents
upstream nodes from hogging the bandwidth, but there is also a
"fairness algorithm" by which the nodes allocate ring throughput to
various users--every node understands the entire ring.
"Fairness" and bandwidth allocation depend on a new Media Access
Control (MAC) layer (yet another header). RPR isn't pure Ethernet
on the ring, but it is fully compatible and has been blessed by the
IEEE (in June 2004) as 802.17. All connections off the rings are
standard Ehernet (unless a vendor has added TDM circuit emulation or
some other feature in a node).
RPR frames travel from source node to destination node, not around the
entire ring (as in Token Ring). As a node removes packets
addressed to it, downstream bandwidth is freed for other users--RPR
calls it "spatial reuse" (drop & insert).
Another function of the RPR MAC is to locate any break in the ring so
the nodes can redirect packets in the opposite direction if necessary
to avoid the break. More than "sort of" like SONET, RPR corrects
for the fault in under 50 ms, the same specification as for SONET.
PROVISIONING IMPACT
Bandwidth allocation on each fiber in an RPR is as flexible as routing
(or statistical multiplexing). The configuration can allocate
guaranteed capacity to some user connections, allow some of them to
burst to higher average throughput, limit some via traffic shaping, and
offer best-effort service to the rest. Since each direction is
independent, high priority connections on one fiber can have reserved
back up paths on the other fiber.
Who gets what depends on standard routing functions: source and
destination addresses, physical ports, DiffServ code points,
application IDs, etc. Every choice is expressed in a software
configuration, not a cable placement or fixed TDM allocation. You
could, for example, envision an RPR node on a dual-fiber ring that
connects to all local devices (adds/drops) through a single GigEnet
interface.
A Local Exchange Carrier (LEC) could find it handy to aggregate all
traffic at a site onto one access link to the central office or point
of presence (POP). What's downstream of that point could be
anything: DSL access mux (DSLAM), cable modem headend, another
router, an Ethernet switch, etc.
Specific services (Internet access, remote storage, email links, etc.)
could be provisioned in equipment removed from the RPR node or part of
that node. RPR offers three levels of priority, so most traffic
engineering schemes map easily to the RPR format. If existing
devices shape traffic, convert Plain Old Telephone Service (POTS) to
voice over IP (VoIP), etc., physically the RPR node could be very small.
Looking ahead to increasing demands for high availability and huge
throughput, imagine a pair of RPR nodes on the customer premises.
Ideally, the "east" and "west" paths for the fibers would be routed
diversely to other RPR nodes or POPs.
With another function of the RPR MAC header, multicasting, those two
nodes would share one RPR MAC address (and by extension, would be
reachable at the same IP address). Then no matter what single
failure occurred on the ring (loss of the local node excepted), the
customer would never lose connectivity upstream.
When the need for bandwidth increases, the same fiber ring will carry
huge amounts by adding wavelengths. Some RPR vendors have methods
to add wavelength division multiplexing (WDM, not to be confused with
WMD) very economically by letting some wavelengths "skip" nodes (the
wavelength is passed through). When traffic on that wavelength
needs to be dropped or added at another node, install a WDM module
there.
RPR supports convergence on IP services with Ethernet compatibility,
quick recovery from faults, and traffic engineering features for
prioritization and bandwidth allocation. The top priority is good
enough to offer low-jitter TDM circuit emulation.
Want all the details? IEEE waits six months to post new 802
standards for free download, so 802.17 should be available that way
around December 2004. Meantime, you can buy a copy from
IEEE. For not-quite-as-detailed information, refer to the many
articles and white papers offered by vendors.
Write if you have a favorite access technology you'd like to see
covered in ViewsLetter: editor@viewsletter.com
============ + ===== + ===== + ===== + ============
PON MARKET REVIEW
For a close look at more than two dozen
PON companies, you are invited to purchase the report, "PON Industry
Players--2004" available from Flanagan Consulting. Offered on
paper and a PDF file via email, this 25-page document describes each
company's products, shows which market segments they participate in,
and provides current contact information. Either form is priced
at US$60.00, payable by check to Flanagan Consulting, 45472 Holiday Dr.
#3, Sterling, VA 20166.
============ + ===== + ===== + ===== + ============
FLANAGAN CONSULTING
-- Call us for a vendor-neutral network architecture and strategy for
expansion or convergence. We know voice AND data--and how to
avoid expensive bear traps on the migration path, such as security
arrangements.
--Working on product positioning or a marketing message for
telecom? Yes, we've done that--for hardware products and
carrier services.
-- Need an Expert Witness? Associates at Flanagan Consulting have
aided in many legal proceedings involving telecom intellectual property
and technology.
--For RFP preparation, bid analysis, proposal evaluation--call
us. We have current experience in Federal network procurement
processes.
"We Have the Experience."
-- Special thanks for supporting ViewsLetter to www.webtorials.com,
your best source for communications tutorials and white papers.
============ + ===== + ===== + ===== + ============
"Flanagan Consulting" and "ViewsLetter"
are Service Marks of W. A. Flanagan, Inc.
|
