
May 1999
A Glimpse Into The Future Of CompactPCI
BY CHRIS DONNER
As someone who receives innumerable press releases, outlooks on the industry, and
positioning statements each day of the work week, believe me when I say that there is
something happening with CompactPCI. More and more frequently I see offerings from all
kinds of vendors - board developers, industrial computer manufacturers, switch designers -
boasting of their adherence to and use of the CompactPCI form factor.
Recently I had the opportunity to travel to the West Coast and take part in the
Motorola-sponsored MediaDay IV. This event detailed many of the recent achievements in the
field of CompactPCI, as well as pointing out some areas that will require more attention
and development in the future. Participants in the event included traditional CTI players
such as Dialogic, Natural MicroSystems, Ziatech, and Force Computers, as well as some
other voices. Notable inclusions were operating systems vendors - both Sun Microsystems
and Microsoft Corporation were present, with Microsoft delivering the opening
presentation. A few days after the Motorola event, Microsoft and Dialogic announced that
Microsoft would be playing a bigger role in the telecommunications space through an
investment in Dialogic and involvement in Dialogic's CT Media. (For more on this
announcement, please see Rich Tehrani's Publisher's Outlook
in this issue.)
A NEED FOR SOFTWARE
The inclusion of operating system vendors is significant. The various presenters at Media
Day IV were pretty much unanimous in stating that the software field is the area where
CompactPCI currently still needs the most work. PICMG (PCI Industrial Computers
Manufacturers Group) and board/chassis vendors have been quite successful thus far in
realizing hardware interoperability and support for hot-swap and high-availability systems
through CompactPCI. However, as Joe Pavlat, PICMG's president, clearly pointed out,
hardware is only one piece in the puzzle.
One important software issue that must be addressed is how the operating system (OS) is
made aware of and compensates for board-level changes, as well as how the OS is able to
handle the use of multiple processors to provide both greater processing capabilities and
failover in case of a CPU failure. This relationship between the hot-swappable components
(including the CPU itself) and the OS is crucial to achieving telco-grade reliability on
CompactPCI platforms.
One company addressing this issue directly is Pigeon Point Systems. Pigeon Point is
working directly with Microsoft and PICMG on the Hot Swap Infrastructure Project to insure
that Windows 2000 (Windows NT 5.0) supports hot-swap in a plug-and-play architecture. This
development process consists of two components: 1) managing of the platform/board hardware
independently of the platform/board vendor, and 2) using plug-and-play to accomplish
hot-swap software connection.
On the hardware side, the core hot-swap driver manages all hot-swappable boards
independently of the board vendor, as long as the boards support the CompactPCI hot-swap
specification. However, some parts of the physical removal process are platform-specific,
such as how the system handles the ENUM# - a status indicator that the central processor
detects when polling available slots and which lets the CPU know the state of the boards
in these slots to provision system resources. On the software side, the goal is to allow
standard, hot-swappable devices with self-contained drivers that will support standard
plug-and-play hardware and Windows NT drivers.
The importance of addressing the software side of hot-swap was also stressed during the
presentation by Sun Microsystems. Sun cited statistics demonstrating that 92% of all
system failures are software related. This accounts for the vast majority of system
downtime. Additionally, while hardware vendors abound, there are relatively few vendors
who have control over system software - and Microsoft and Sun are two of the biggest
players in this area.
CLUSTERING
Sun proposed that a possible way to address the software failure issue was by using
clustering to gain the benefits of fault-tolerant systems without incurring the associated
expenses. Additionally, clustering allows more flexibility in deciding which operating
system(s) to run on a system. Both clustering in general and the operating system
flexibility are key differences between Sun's strategy and Microsoft's. Unix is generally
perceived as offering more advanced clustering capabilities than Windows NT, although
Microsoft is working hard to alter this perception, and the flexibility to use various
operating systems across a cluster mean that a carrier or service provider won't be locked
into using only Windows or only Unix across a particular node.
Clustering alone isn't sufficient to allow for true fault tolerance in a network,
however, and there are a few important conditions that must be met. Among others, Sun
cited the need for: dynamic addition/subtraction of node elements and nodes themselves,
high-availability foundations and parallel extensions for standard databases, cluster
specialization, and remote global management facilities. Some future challenges Sun
pointed to for the CompactPCI development community include: high-speed clustering
interconnects, leveraging existing closed cluster environments, and reducing clustered
domains to a single backplane.
Force Computers also addressed the issue of clustering to achieve high-availability
systems. A key question asked by Force was: When do I detect the failure in my system?
Given an early enough detection - prior to the failure affecting the application -
failover can occur more easily, and recovery can take place more quickly. A
high-availability cluster system would be suitable for database and network management
applications, allowing failover times that average 30-60 seconds per incident. But carrier
grade, five-nines reliability would require host hot-swap features, an area where a great
deal of software work is yet to be done.
CompactNET - ONE ALTERNATIVE
Ziatech Corporation also had some ideas regarding increasing system performance and
allowing for modularity by using multiple systems, although instead of clustering, they
focused on multiprocessing. The idea behind Ziatech's CompactNET architecture is
interesting. Utilizing the PCI resource arbitration inherent in the CompactPCI bus
architecture, CompactNET allows multiple processors to be linked together over the
CompactPCI bus, which the processors perceive as nothing more than an Ethernet connection.
The processors are joined in a "multiple-master" configuration, in which each
processor is responsible for driving its own data/control signals to the backplane. In
this manner, CompactNET allows for full-mesh connectivity between system processors - any
CPU can communicate with any other CPU in the backplane without involving an external
"master."
The benefits of Ziatech's multiprocessing approach are significant: multiprocessing
systems are generally faster and easier to develop on than single-processors systems
running multi-threaded, multi-tasking operating systems. Maintenance and upgrades to
multiprocessing systems should be more straightforward than on dedicated systems.
Additionally, the bandwidth available to CompactNET systems increases as nodes are added.
And finally, it is beneficial to be able to use multiple operating systems. For example,
an application might be run more effectively with a real-time operating system, while the
interface for the user might be more functional if provided through a GUI on a
general-purpose operating system.
OPEN SOURCE CompactPCI
Readers of Brough Turner's "On The Horizon" column in CTI will
already be familiar with his announcement of Natural MicroSystems' Open Source for Open
Telecom initiative. The announcement of this initiative was initially made at Media Day
IV, and it was quite significant for the various players in the field of CompactPCI.
The www.opentelecom.org Web site includes not
only information about CompactPCI but actual source code for such things as Natural
MicroSystems' CT Access hot-swap infrastructure, point-to-point circuit switching service,
and sample device drivers. NMS has released this code under Open Telecom Public License,
which derives much of its language from the Mozilla Public License (www.mozilla.org) that Netscape Communications
Corporation used in making public the source code of the Communicator 5.0 Standard
Edition. (For more information on open source, visit www.opensource.org.)
NMS took this rather surprising step in an effort to drive growth in CompactPCI market
for the telecom market. The idea behind the release is that a booming market will be more
profitable for NMS (and other vendors) than a slowly developing market would have been had
NMS kept their source code in-house. Additionally, the open source model is meant to
promote interoperability and use of standards, since other board vendors might easily
model their own drivers and coding on NMS's working code.
As of March 4, PICMG voted to endorse the Open Source for Open Telecom initiative. In
addition, the PICMG executive membership voted to authorize the officers of the
organization to bring the Open Telecom initiative into the PICMG sphere of activities, as
a model for future software enhancement efforts. "The Open Source model and the
specification model both have their places in PICMG," said Joe Pavlat. "They
will complement each other in providing both definition and implementation details."
SUMMARY
Overall, the message at Media Day IV was quite positive. There is a sense of
accomplishment among the participants, in that PICMG has released, and the vendor
community has responded to, several standards defining critical hardware issues for
CompactPCI and hot-swapability to become a reality. There is also a sense of challenge and
excitement regarding the much needed software solutions that will allow CompactPCI to
provide the true five-nines reliability needed in the telecom space. With continued focus
on software specifications and solutions that build on what has already been accomplished
on the hardware side, CompactPCI is clearly becoming a major force in the
telecommunications industry.
Chris Donner is associate editor of CTI magazine. He can be reached for comment
at lguevin@tmcnet.com.
|