(Barely) managing ATM
Where the once and future network king
ATM attracts attention as a technology that solves not only tough network problems of today, but promises the speed and flexibility to answer demands of the future as well. ATM standards and tools emerge slowly, however, and many challenges in ATM network management remain. (2,600 words including two sidebars)
Only a year ago, ATM (Asynchronous Transfer Mode) was considered the strategic solution for any networking or communications need. No respectable industry watcher or network planner could imagine a future without ATM, and few foresaw network architectures, either for LANs or WANs (local or wide area networks), beyond it. ATM products and standards, however, are emerging more slowly than predicted. The ATM explosion may detonate a few years later than everyone expected.
According to market researcher Vertical Systems Group (Dedham, MA), sales of Frame Relay (one of ATM's logical parents) access equipment will more than triple in the next three years from $171.8 million in 1995 to $613.4 million by 1998. ATM takes this popular WAN technology one step farther into cell relay, and therefore is expected to pick up where Frame Relay left off. A cell is a fixed-size, short frame. In the case of ATM, the cells are 53 bytes long.
Vertical Systems Group estimates the number of customer organizations paying for public ATM WAN services at the end of 1995 numbered no more than 72. The market research firm expects this number to grow to about 1,000 by the end of the century. (See the sidebar ATM network technology 101 for more background information.)
What the future holds
Only a few pundits challenge the notion of ATM's eventual dominance in WANs. ATM's future for LANs, however, is less certain thanks to the blooming Fast Ethernet and the budding Gigabit Ethernet. According to market research firm IDC, low cost and compatibility with the existing LAN infrastructures helped Fast Ethernet adapters outsell ATM adapters by about 20 times in 1995, when 665,000 10/100Base-T adapters were sold. ATM was a distant fourth, with 38,000 adapters, trailing 100VG-AnyLAN's 218,000 and FDDI's 124,000.
"The dream of an all-ATM network is history," said Paul Callahan, director of networking research at Forrester Research (Cambridge, MA). "Users are demanding more raw bandwidth, not the real-time services of ATM cell switching. ATM will remain a niche technology used mainly in the backbone of the network or used by carrier networks."
Some users don't agree. Mark Spencer, director of networking for the University Hospital in Denver, CO, provides physicians in the critical care unit with instant data access using ATM networks. "We went looking for a solution, not technology. ATM just happened to be part of the solution. Many vendors, carriers, and users believe in the future of ATM both on the LAN and WAN and continue to develop not only standards, but also new products for ATM."
One of the most notable products in this area comes from Sun Microsystems. The recently announced SunATM 622 adapter is the fastest ATM end-station adapter available. The SunATM 622 promises 622-megabits per second network bandwidth over multimode fiber. It provides additional performance advantages over the previously released ATM 155 adapter that supports 155 megabits per second over multimode fiber and Category 5 UTP (Unshielded Twisted Pair). SunATM adapters, which are SBus cards, use ATM SAR (Segmentation And Reassembly). Sun promises the cards are compliant with existing and newly developed ATM Forum standards.
In addition to being the fastest ATM end-station adapter today, SunATM adapters support more users and multiple logical addresses on a single physical link, allowing individual stations on a LAN to appear to be multiple stations. This feature is particularly useful for backup/recovery and server consolidation. A station, using a SunATM adapter, can be a member of up to 16 emulated LANs. The SunATM adapters also offer advanced management functions, allowing ATM SNMP agents to simultaneously co-reside on the same machine with the non-ATM SNMP agents.
ATM network management: What's available and what's promised
One of ATM's major challenges is network management. The five-layer network management model created by the ATM Forum provides a structured approach for managing ATM networks. (See the sidebar What's the ATM Forum? for more.) The Network Management Working Group of the ATM Forum is finalizing this end-to-end management model that will be applicable both for private and public networks. The model will also define gateways between SNMP management systems (e.g., SunNet Manager), CMIP (Common Management Information Protocol--see Resources for more), and the existing proprietary systems. It will also define special OAM (Operations, Administration, and Management) cells to distribute management information over ATM networks.
There are five key management interfaces, labeled M1 to M5, defined in this framework.
A special status in this framework is dedicated to SNMP--the de facto (and de jure) standard for network management. M1 and M2 interfaces encompass the relevant standard MIBS (management information bases) for DS-1, DS-3, SONET (Synchronous Optical NETwork), and the AToM MIB--as defined by RFC 1695. The AToM MIB allows network managers to group collections of switches, virtual connections, interfaces, and services into discrete entities and is SNMPv1- and v2-complaint. It is expected to reduce the proliferation of existing proprietary vendor-specific ATM MIBs.
Since most LANs today are SNMP-manageable, and carriers are still using CMIP (as well as proprietary network management protocols in many cases), the ATM Forum recognizes that private and public network management systems will maintain separate views into the network for some time. Therefore, ATM Forum's network management scheme will enable both views and involve the building of gateways between these dissimilar systems.
Other groups, such as the Network Management Forum and the ITU (International Telecommunication Union), have been working with the ATM Forum to negotiate acceptable interworking standards. Obviously, it will take time to finalize specifications for interworking between SNMP and CMIP. The good news is that some progress has been made with the M4 interface toward establishing a protocol-independent MIB, which will support SNMP objects based on SMI (Structure of Management Information), as well as CMIP objects, conforming to GDMO (Guidelines for Development of Managed Objects). The bad news, however, is that there is a long way to go for full bi-directional interoperability between SNMP and CMIP.
Within the next few years, one may expect many developments in ATM management. However, at least in the earlier stages, most developments are likely to be based on OAM cells, rather than on SNMP or CMIP architectures. The ATM Forum is currently specifying three types of 53-byte OAM cells with specialized identification and field tags, to identify their functions: fault management, performance management, and activation/deactivation (for starting and terminating fault and performance management functions). By providing ATM network devices with the ability to gather information about end-to-end connections, the OAM cells will reduce the need to distribute MIBs throughout the network, and significantly lighten the amount of management-related traffic.
Standards for managing ATM performance
Efficient network management is one of the major functions required in any networking architecture. What tools do we have today for managing ATM networks? RMON (Remote MONitoring) was developed for shared media (such as LANs) and does not scale well to switched architectures (such as ATM). RMON, as well as protocol analyzers, need an ATM probe to act as a passive observer, which can unobtrusively steer cells to where they can be reassembled into packets for debugging.
In July 1995, a group of manufacturers of ATM switches, test equipment, and RMON products announced formation of the AMON (ATM MONitoring) group to develop an AMON agent, which will route copies of troublesome virtual circuits to a test port, for monitoring ATM layers. While some ATM-specific MIBs already exist, the proposed ATM Circuit Steering MIB expands the contents higher in the ATM stack to include virtual paths, virtual circuits, destination ATM addresses, and timers. AMON capabilities and objectives of the ATM consortium include:
However, additional standards are required to define what data should be collected and how to present that data to a network management system. An interesting development in this arena recently came from a major player, Cisco Systems. Cisco harbors major plans for ATM technology in light of its recent acquisition of Stratacom for a cool $4 billion. In February 1996, Cisco, with a group of RMON and analysis tools manufacturers, developed a new draft for ATM remote monitoring. This draft defines the means of how an RMON agent, embedded on ATM switches and distributed across ATM networks, can gather network traffic information. It is based on the original RMON (RFC 1271) and RMON2 (RFC 1757) specifications.
Interestingly, Cisco, one of ATM Forum's founders, submitted this draft to the IETF (Internet Engineering Task Force) rather than the ATM Forum. Cisco and six other vendors plan to submit it to the ATM Forum later this year. This ATM RMON draft defines different methods that can be used to collect network traffic information. One of the methods uses "circuit steering" as proposed by the AMON consortium. In addition to circuit steering, the draft defines the option of having cell-level RMON instrumentation integrated in an ATM switch fabric. To size up ATM traffic use, the draft employs basic cell statistics per port, per virtual connection, per host, and per conversation pair.
Such detailed information can help network managers determine the performance bottlenecks and find the top talkers in an ATM network. It is quite challenging to capture the data at full rate merely because of ATM's 150-to 650-megabit per second fabric and the limited amount of available storage. Therefore, the specifications also discuss data sampling methods that allow capture of frames, rather than random cells, which would yield little useful information.
ATM management tools to the rescue
As we can see, management of ATM networks is quite a challenging task. Since most of the standards are still under construction today, your best bet is to use vendor-proprietary tools specifically designed for their hardware, and address ATM network management with carriers and public service providers as early as possible during contract negotiation. It is probably a good idea to request a direct feed from the carrier's network management system into your network management center. It is always wise to monitor their capabilities, and, obviously, their prices. Although the list of ATM public service providers is still quite short, one can now find among them not only the four major long distance carriers (AT&T, Sprint, MCI, and LDDS Worldcom), but also Bell Atlantic (Bedminster, NJ), Brooks Fiber Communications (Town and Country, MO), MFS Datanet (San Jose, CA), Pacific Bell (San Francisco, CA), and US West (Englewood, CO).
Most vendor-proprietary tools for ATM network management will run on the major SNMP network management platforms, such as SunNet Manager. Notable tools include:
There are also hardware-based tools and equipment that fall mainly into two categories:
The first category of high-end ATM analyzers usually comes with advanced traffic generators that can capture and replicate ATM activity correlating to several PVCs.
Examples of such products include the AX/3000 and AX/4000 ATM Cell Generators from Adtech (Honolulu, HI), the ATMax-200 from RAD Ltd. (Tel Aviv, Israel), which tests and generates ATM traffic from T1 or E1 (1.5 or 2 megabits per second) all the way up to OC-3 (155 megabits per second), and CellBlaster from Net2Net (Hudson, MA) that can support up to 32 ATM trunks and can filter, capture, and perform ATM-layer decoding at full-bandwidth automatically, full-duplex for up to OC-3 speeds.
Such analyzers are well suited and equipped for stress-testing ATM switches and networks. CellBlaster and ATMax-200 can also be ordered with non-ATM interfaces, such as Ethernet, Token-Ring, FDDI, and others. Net2Net recently added an important new feature to CellBlaster--an SNMP agent. The addition of the SNMP agent made CellBlaster visible to SNMP-based network management systems, such as SunNet Manager, allowing it to be displayed as a network entity on the management console and respond to SNMP queries with statistics pertinent to the network's health. The product currently supports the SONET/SDH (Synchronous Optical Network/Synchronous Digital Hierarchy); (RFC 1595) and DS3/E3 (RFC 1407) MIBs for monitoring physical layer interface statistics.
Future MIB support will be expanded to include:
Most products in the second category can also generate and analyze ATM traffic. However, they usually can't create the same traffic scenarios, and some of these tools are limited to lower speeds than available on the high-end ATM-only analyzers.
Managing ATM networks -- Is it possible?
Managing ATM networks with virtual connections at a much higher speed is substantially more difficult than shared-media LAN networks. To make it even more challenging, ATM network managers are required now to administer multimedia networks, carrying not only data, but also voice and video. With ATM, they have to manage and fine-tune four different QoS (Quality of Services):
SNMP was not designed to handle the titanic volumes of network traffic ATM generates. SNMP-based tools have been extended to support ATM networks. To accomplish this, network management architecture for ATM switched networks is decentralized and deployed as a series of agents that report from various ATM devices when a connection is established. By using this approach, ATM devices can send SNMP data to a central network management station, thus providing SNMP monitoring in real-time. It also extends the role of network hardware. ATM hubs are now playing a critical role in network management strategy. One of the most important conditions in selecting an ATM hub is to make sure it supports a multi-agent SNMP architecture to allow reconfiguration without losing its manageability.
The emerging high-speed networking technologies will provide users with much more flexibility in solving complex network problems. Networks can now be designed to support concurrent data, video, and multimedia applications at mega and even gigabit speeds. Among the emerging technologies a special place is reserved for ATM.
Despite the lack of fully developed standards today, ATM is becoming extremely popular with most hardware vendors and carriers. While some bumps on the road to ATM were uncovered recently, it is still considered the most promising communications technology. Since management of ATM networks has the highest priority, not only for users, but also for vendors, carriers, and standard organizations, all hope ATM will become a manageable technology.
If you have technical problems with this magazine, contact firstname.lastname@example.org
Pundits and users consider ATM the strategic solution for future networks. One of ATM's major advantages is its ability to carry voice, data, and video. Here are some of the main reasons users consider migration to switching technology in general and ATM in particular:
An additional advantage in the eyes of many (but disadvantage in the eyes of few, recently becoming more vocal) is the fact that ATM is being developed by a consortium of more than 750 members -- the ATM Forum. As with any standard developed by industry consortia with many vendors, some advancing their own agendas, adoption of ATM is not progressing rapidly. Although many difficulties have been overcome, and many ATM standards have been defined and turned into products, ATM still faces a long road ahead.
Based on small, 53-byte packets (or cells), ATM works well for traditional data processing applications (which can sustain delays but no loss of data), as well as for voice and video applications (that can sustain some loss of data but no delays). ATM technology uses the same uniform short-frame format on LANs and WANs. For example, a frame cell generated by a 100-megabit per second LAN can be carried over a 45-megabit per second DS3 circuit and switched onto a 2.4-gigabit per second SONET transport system. During its journey from a LAN via a DS3 WAN to a SONET transport system, the frame format does not change. Rather than implementing complex protocol conversions between LANs and WANs of dissimilar nature, ATM involves address manipulation and flow control, when used in both environments.
Merging the best features of TDM (time-division multiplexing) and packet switching, ATM provides a versatile, multifunctional platform that can support a variety of services and traffic types. In addition, ATM also offers scalability that is difficult to match with other conventional solutions. The short size of the frame, which is a compromise between the long data formats in classical IT applications and short, repetitive pattern of voice-and video-intensive systems, allows mixing of all three data types within the same application over the same network adapter. One of ATM's major advantages is that it allows designers to tailor the network to the application, rather than forcing the application to fit network limitations. It supports delivery of traffic at rates and burstyness compatible with the application.
ATM is an OSI layer 2 protocol, specifically designed for hardware-based switching of short fixed-length packets over high-speed links across large distances. One of the most important features of ATM is delivery of packets in the order they were placed into the network. An ATM cell may encounter congestion and experience variable delay, and it can even be dropped because of a transmission error or be dropped by a congestion control mechanism. However, cells with the same VCI (Virtual Circuit Identifier) are always delivered in the order in which they were submitted. Because there are no re-transmissions and no reliable delivery within an ATM network, higher layers of the OSI protocol stack have to take care of end-to-end acknowledgments and guarantee delivery of data. One of the most comprehensive collections of Web sites with information on ATM technology, concepts, and products was compiled by Syracuse University.
As a still-evolving architecture, the size of ATM cells was a compromise between the European proponents of 32 bytes and American backers of 64-byte cells. The parties met halfway and agreed on 48-byte frames (with 5 additional bytes for control information).
Formed in October 1991 by Sprint, NET/Adaptive, Cisco, and Northern Telecom, the ATM Forum currently includes more than 750 companies representing all sectors of the communications and computer industries, and a number of government agencies, research organizations, and users. One of the misfortunes some standards committees suffer is the perception of the Unfinished Symphony syndrome thanks to their perpetual drive for improvement. ATM Forum is no stranger to this phenomenon.
To advance the completion of ATM standards and regain momentum for ATM acceptance, the ATM Forum decided to prioritize its tasks. In its Anchorage Accord formulated during an April meeting this year, the ATM Forum defined Foundation Specification--the baseline fundamental for scalable and interoperable ATM implementation, and Extended Specification, which has a lower priority.
Foundation specifications consist of:
Extended specifications include:
To interpret the Anchorage Accord, Steve Walter, the president and chairman of the ATM Forum said: "It has been suggested, that this step will put the ATM Forum on a holding pattern or into early retirement, thereby causing the Forum to lose it effectiveness. But this is not the intention, nor will it be the result.... We feel the time is right to communicate to the industry that a cohesive set of specifications exists for providing the ATM infrastructure foundation and service foundation. This will enable service providers and suppliers to base their implementations on interoperable versions of ATM Forum specifications. Users can then migrate to ATM services and systems without delay. So is ATM ready for prime time? We believe the answer is definitely yes."
About the author
Eddie Rabinovitch is a senior manager with the Network and Desktop Consulting Practice, Global Customer Services at Unisys Corp. He has published more than 50 papers with several trade and technical publications and is a frequent speaker at national and international conferences. Check out the Unisys site at http://www.unisys.com Reach Eddie at email@example.com.