"You say 't-m-t', I say 't-mh-t', let's
call the whole thing off." Or in this case, you say,
"CompactPCI" and I say, "switched Ethernet
fabric" or "StarGen fabric" or "Infiniband."
But Ira Gershwin lyrics aside, hopefully we won't be calling
the whole thing off anytime soon. In fact, CompactPCI, which
began life as an industrial form factor back in 1994, has
proven itself a serious player in the telecom market, and it
is this seriousness that is redefining that form factor in a
way that makes me wonder whether it will still even be
called CompactPCI (cPCI) once the changes are complete.
Anyone familiar with the rise of cPCI over the past six
years or so will know cPCI's value proposition: Take
advantage of the mass market PCI form factor while improving
its reliability for industrial-style applications, including
telecom. Since 1994, telecommunications has arguably become
the key space for cPCI deployments, with around a billion
dollars worth of shipments in 2000 (depending on which
figures you use).
But densities in the central office are rising to meet
increasing traffic demands due to growth of the wireless
market, longer connection times for dial-up Internet access,
and development of enhanced services tying together voice
and data applications. Correspondingly, cPCI must meet these
density needs or see itself eclipsed by other form factors
that are perhaps less open but that save carriers and other
service provider valuable space.
The PCI Industrial
Computer Manufacturers Group (PICMG) has several
projects underway to address these issues of system density,
and I recently had a chance to speak with Joe Pavlat,
president and chairman of PICMG, about some of these
initiatives and the future of cPCI in general. Afterwards, I
also talked with John Peters, chairman of the PICMG 2.16
subcommittee, the working group focusing on the integration
of switched Ethernet fabric into the CompactPCI backplane,
creating what is known as the CompactPCI Packet Switched
Backplane (CompactPCI/PSB, or CompactPSB).
Rumors and realities of a threatened recession aside,
carriers are mainly concerned with one thing: Density. More
calls per shelf mean lower real estate costs and easier
network management, among other things. As Joe Pavlat
pointed out, a 6U cPCI board has 535 pins, only 220 of which
are devoted to the existing PCI bus. This leaves 315 free
pins to be used as a manufacturer sees fit. As a result of
increased demand, many cPCI vendors have already begun
experimenting with alternative data buses running in tandem
with the H.110 bus in cPCI.
I was able to get an overview of progress being made in
PICMG on three fronts to address issues of increased
densities in the cPCI chassis. These initiatives include the
aforementioned 2.16 workgroup, bringing an Ethernet pinout
directly into the cPCI backplane; the StarGen workgroup,
which utilizes a bridge chip to link the PCI bus to a switch
fabric; and Intel's Infiniband architecture.
The introduction of Ethernet into the cPCI backplane is very
much in keeping with the original philosophy behind the
development of the cPCI form factor itself: Take existing
technology and adapt it for industrial purposes. This
philosophy has many advantages, including the ability to use
off-the-shelf (OTS) components, to benefit from widespread
implementation and debugging, and a familiarity on the part
Ethernet is certainly a well-established technology, and
it is also inherently hot-swappable, something that is
absolutely required in high-availability telecom systems.
Additionally, Ethernet is an open specification and allows
for communication between systems running different
operating systems, so it is possible to imagine a chassis
containing multiple boards, each of which has its own
memory, CPU, and OS -- all of which can still communicate
with each other quickly and effectively using TCP or UDP or
John Peters argued that these advantages are more important than the purely
technical merits of a particular specification, and this is a powerful point.
Ethernet is here and it's well known: It has already achieved critical mass. And
its capacity is increasing as well, from 10/100 to gigabit and soon to 10
gigabit. As John put it, "We're simply putting the pieces together in a unique
Unique... and popular as well. Joe Pavlat informed me
that in terms of the number of meetings and the number of
participants, the PICMG 2.16 committee is the biggest PICMG
committee so far. Many of these participants are also ready
to support the spec with OTS components once it's ratified,
which John Peters says should happen around the end of July
of this year, perhaps a bit later. Notably, this is less
than a year after the first meeting of the subcommittee,
back on November 8, 2000.
Baby, I'm A Star...
So that's it then, right? When the dust settles, it's
CompactPSB from now on? Well, not quite. StarGen
and other members of the StarGen workgroup have implemented
the StarGen point-to-point, high-speed switch fabric into
the cPCI form factor in an attempt to support speeds up to
OC48, and calls in the 10s-100s of thousands per chassis.
The system employs an H.110 bridge chip that integrates
the H.110 backplane into StarGen's high-speed switch fabric.
The bridge chip translates H.110 traffic into the
StarProtocol, allowing full backwards compatibility with PCI
and H.110 and allowing a flexible adoption rate on the part
of carriers, but also supporting high-speed, QoS-based
switching of voice and data traffic.
Systems can be linked directly, using only a bridge-only
configuration, or can be linked through the StarGen switch
for large configurations, using additional switches for
redundancy and additional capacity. Connections can be made
on a chassis-to-chassis or point-to-point basis, as well as
allowing for segmentation of the bus into multiple
independent buses. Joe assured me that the StarGen
technology will be an open specification as well, again
assuring rapid and widespread usage and deployment.
One of the real strong points of the StarGen solution is
its ability to handle TDM and isochronous traffic, which is
important for carriers looking for a current solution that
will scale and allow them to move over to an IP-based
network without obsoleting network equipment along the way.
This is an important consideration when you remember that
fully replacing the current PSTN with an all-IP network is
estimated to be a 10-15 year process at the minimum.
Both Joe Pavlat and John Peters are also involved in the
development of Intel's
Infiniband architecture, but they both also expressed
reservations about this process. Most prominent in their
concerns was the newness of the technology and the length of
time before it would be ready for deployment.
John specifically referred to the long gestation period
required for a new technology, citing the example of flash
memory versus rotating memory. Despite the availability of
flash and its technical superiority, most of us still use
rotating disk drives in our PCs and laptops. Also, John
suggested that if you project where current technologies
like Ethernet and StarGen are likely to be when Infiniband
is ready for use, the benefits of Infiniband in a telecom
environment are less clear. Again, this is similar to the
flash situation, since you can now go to the local computer
megastore and easily and cheaply pick up a 20GB drive using
rotating memory -- something not necessarily expected when
flash was initially developed.
Joe Pavlat also questioned whether Infiniband really
solved any problems that won't already be solved by gigabit
Ethernet or optical (10G) Ethernet. Without clear
performance or time-to-market advantages, Joe suggested that
Infiniband will be useful in the commercial server world but
that it is somewhat less compelling in the telecom space. It
will certainly be interesting to watch the advances made
with Infiniband, both within and outside of the central
office space, but John and Joe make strong points supporting
both the StarGen fabric and the CompactPSB Ethernet
alternative, and at the moment these two options appear to
be the most promising.
Let's Call The Whole Thing...?
We've looked at options for significantly increasing cPCI
densities over the next few years, but I'm an editor after
all, and words are important to me. Much like they were to
Ira Gershwin. Will the new cPCI still be called CompactPCI,
or will some other name perhaps become more popular and
appropriate? I asked Joe Pavlat to speculate along these
lines, and he laughed a bit and then pointed out that the
original working name for CompactPCI was actually
"rugged PCI," but that for various reasons this
name was changed during development.
Still, if traffic is eventually being carried over
high-speed secondary buses, and the PCI bus becomes more of
a command and control bus, then why preserve the PCI name?
What's in a name, after all, right? Well, marketing for one.
Familiarity also -- look at all the changes that have
happened with Ethernet over the years, and yet it's still
called Ethernet. If the next few years sees cPCI
implementing all kinds of changes -- integration with both
StarGen and Ethernet, perhaps integration with Infiniband,
hybrid systems using two or three of these technologies or
even proprietary solutions -- then perhaps the simplest
thing is to continue using the CompactPCI name, even if it
isn't 100-percent accurate any longer.
One certainty is that even the originators of the cPCI
form factor have been pleasantly surprised at the speed and
extent of its acceptance into the telecom market. And they
are likely to agree that whatever the final name, there's no
need to be calling anything off. Since we're on the subject
of names and words, I'll steal a recently coined word from
our bumbling president: Don't "misunderestimate"
this technology. No matter how you say it, no matter how it
achieves higher densities, CompactPCI is planning on being
around for a long time.
is a contributing editor to Communications Solutions
magazine and editorial director of eNews In Your Inbox.
To The June 2001 Table Of Contents ]