Bringing ATM into your Network A Review of the Integration Options --------------------------------------------------------------------------- Introduction Many network managers are now making plans to integrate ATM into the network. While ATM promises to increase application performance, the truth is that achieving greater network performance is not as simple as installing ATM switches, routers and adapters. Some hardware vendors may maintain that this type of migration has grown increasingly simple, yet there are still a number of issues that must be understood. Adaptec provides this document to explain these issues. This information should serve as a valuable guide, detailing some of the ATM-to-legacy system integration options now available. The challenge before you. In most scenarios, existing legacy networks rely on Ethernet, FDDI, token ring, or some combination of these protocols. Integrating ATM connectivity solutions, at least for selected applications, is the next step. However, an enterprise-wide upgrade would be costly. The challenge is to make the transition as cost-effective and painless as possible -- to maximize the value of the organization's network. Network managers who are planning to integrate ATM usually aim to accomplish one or more of the following: * Inject high-bandwidth connections between routers * Connect small workgroups of high-bandwidth ATM users * Replace the existing backbone infrastructure with ATM, connecting campus servers to ATM, while leaving the legacy clients untouched * Establish a parallel ATM network for high-bandwidth users which can later be expanded to provide service to the rest of the network * Implement a parallel ATM network for inter-server communication --------------------------------------------------------------------------- Choosing an ATM Internetworking Standard Realizing the objectives listed, and gaining the performance boost ATM can provide, requires more than simply installing ATM hardware. Existing applications won't run over ATM without first complying to the internetworking standards that connect ATM to the network structure. So the process begins by selecting the ATM internetworking standard that best meets the needs of the organization. Two such industry ATM standards exist: 1. LAN Emulation (LANE) 1.0, developed by the ATM Forum 2. RFC-1577, or Classical IP (CIP), developed by the IETF These standards function to: * Enable existing applications to run transparently over ATM with no modifications * Enable ATM hosts to transparently talk to Ethernet and token ring hosts To create ATM networks, parallel networks, and backbones, network managers need to first select either the LANE or CIP standard. In doing so, two factors must be evaluated: 1. The existing protocols currently in the corporate network 2. Performance requirements Where multi-protocol situations exist, LANE is the preferred choice. LANE provides multi- protocol and broadcast support. However, where multi-protocol support isn't required, CIP may be chosen. To accommodate multiple protocols and also provide peak performance, running LANE and CIP concurrently is another option. In these types of network configurations, where both current ATM standards are put in place, the ATM hosts run LANE. A limited number of bandwidth-hungry systems can be networked with the use of high speed CIP links. --------------------------------------------------------------------------- LAN Emulation (LANE) The relationship of LANE to existing LANs. With LANE, ATM systems and clients work together to make ATM switched virtual circuits behave like a legacy LAN segment (i.e. broadcast domain). This allows the broadcast messages of any member of the emulated LAN to be received by all other members of the emulated LAN. LANE's advantage over traditional LANs is that the network segmentation can be done with little regard to physical wiring. In fact, emulated LANs are a standardized form of virtual networks. Emulated LANs are achieved by replacing the Data Link Layer (MAC sublayer) of both Ethernet and token ring networks. Host data handed down to the MAC sublayer is intercepted by the LANE software running on that host. In the receive direction, data received by the host ATM interface is handed up by the same LANE software. With this type of LANE-based solution, higher level protocols don't "perceive" any difference in the MAC layer, and so applications take advantage of the higher speed ATM pipes -- and run transparently. [Image] Figure 1: Broadcasts over LANs The shaded area shows the broadcast domains of a traditional repeater-based LAN versus two emulated LANs in a simple ATM switched environment. Like the traditional repeater- based LAN, members of the same emulated LAN can't be on opposite sides of a router. Component Location in the Function Network In every ATM host and Emulates the MAC layer, LAN Emulation every device attached intercepting all transmission Client (LEC) to the ATM network data and passing up received (e.g. ATM edge device). data. In a switch or in a LAN Emulation stand-alone host. May LEC may talk to the LECS at Configuration be co-located with LES boot time. The LECS directs the Server (LECS) and BUS.Optional LEC to the appropriate LES (see component of LANE. LAN Emulation Server). LES automatically maps MAC Either in a switch or addresses to ATM addresses. LAN Emulation in a stand-alone host. When a client needs to transmit Server (LES) May be co-located with to a specific MAC address, the LECS and BUS. LES notifies the client of the ATM address of the target host. Broadcast and Either in a switch or Handles all broadcast traffic Unknown Server in a stand-alone host. and traffic destined for (BUS) May be co-located with unknown ATM addresses. LECS and LES. Table A: The Components of LANE A LANE environment includes four key software components. Understanding each component will help to better understand LANE as well as the sample network configurations found in the appendix of this document. --------------------------------------------------------------------------- Classical IP (CIP) The relationship of CIP to existing LANs. Like LANE, the purpose of CIP is to allow existing TCP/IP applications and legacy hosts to work with ATM hosts. Unlike LANE however, CIP works with the Network Layer (i.e. the IP layer in a TCP/IP network) instead of the Data Link Layer (MAC sublayer) of the protocol stack. As the name implies, CIP only relates to IP protocols. Therefore, CIP doesn't apply to non-IP protocols such as IPX, DECnet, Appletalk, SNA, etc. Additionally, the current version of CIP does not include support for broadcast or multicast. And connections to legacy hosts can only be achieved through a router. CIP has lower software overhead than LANE, so it is able to move data at slightly higher speeds. Data being sent down to the IP layer is handled by CIP. The data the network interface card receives is handed up from the CIP driver to the normal IP layer. The result is up to 15% higher application performance as compared to LANE. Yet before opting for this approach, a network manager will want to consider CIP's inherent limitations. [Image] Figure 2: CIP talks to legacy network This diagram shows how an ATM CIP workgroup uses a router to communicate with the legacy network. CIP is best suited to small ATM workgroups or server-to-server implementations where peak performance is the most important requirement. Component Location in the Function Network Classical IP In every ATM host. Intercepts all transmission data and client passes up received data. Works directly with the ATM hardware, bypassing the normal MAC layer. ARP server In a stand-alone Provides IP address to ATM address host, or may be resolution. in a switch. Table B: The Components of CIP There are only two components to the current version of CIP (IETF RFC-1577). Lower software overhead results in slightly higher application performance than with LANE. --------------------------------------------------------------------------- Choosing an Internetworking Services Vendor After selecting a LAN emulation standard, it's time to compare vendors. Three significant factors should be considered when evaluating the hardware they offer: 1. Interoperability 2. Reliability 3. Scalability FACTOR 1: Interoperability -- and adherence to standards. LAN emulation standards specify rules that allow products from several vendors to interoperate. However, in reality, implementation quirks may occur. For this reason, demand that prospective vendors perform interoperability testing in private and public forums and can ensure their devices will work with products from their competitors and partners. It's a good practice to ask prospective vendors to provide switch-to-switch, NIC- to-switch, and LEC-to-LES interoperability lists for their products. List of Compatible Switch Vendors List of LANE Vendors ATM Limited Bay Networks Cabletron Bay Networks Cisco Cisco Digital Fore First Virtual Interphase Fore Olicom IBM Whitetree Newbridge Efficient UB Networks IBM Whitetree 3Com Table C: Adaptec means compatibility Adaptec's business model mandates compatibility with industry standards. Our ATM products adhere strictly to approved ATM Forum and IETF standards, such as UNI 3.0 and 3.1, CIP, and LAN Emulation V1.0. Adaptec constantly tests its network interface cards with all the major ATM vendors. Listed above are the switch vendors, followed by the LANE vendors that have been proven interoperable with Adaptec's ATM NICs. FACTOR 2: Reliability -- the lack of redundancy must be resolved. Classical IP suffers from a lack of redundancy in the ARP server and does not support broadcasting. The current version of LANE also lacks redundancy in the LECS, LES, and BUS (Broadcast and Unknown Server). Although these deficiencies will be addressed by in-process standards, they are now being addressed by the industry's equipment suppliers. Component Effects of Component Failure Remedies The LECS vendor can LAN Emulation An LEC that boots up after the offer automated fail Configuration LECS has failed no longer has a over in the LECS. Server (LECS) way to find its emulated LAN Also, the LEC can (i.e. LEC can't find its LES). store the address of its last known LES. The LES vendor can offer automated No new connections can be fail-over capabilities LAN Emulation established on behalf of LECs. to a redundant LES. Server (LES) Existing connections are Also, the LECs can unaffected. cache frequently addressed hosts and fall back to those addresses. Broadcast traffic for the emulated LAN ceases. Additionally, data will not immediately flow between two Broadcast and endpoints in the emulated LAN The LES vendor can Unknown Server which are newly establishing a offer automated fail (BUS) connection to one another. They over to a redundant will have to wait until their BUS. connection is fully established (i.e. the BUS sends data between two endpoints until their connection is fully in place). Table D: The importance of redundancy in LANE services This table highlights the action vendors take to remedy component failure problems. Note that all these components are frequently located on the same host, making redundancy particularly important. You should choose a LANE vendor that can demonstrate a commitment to reliability, including redundancy of the LANE services. FACTOR 3: Scalability -- Are you prepared to grow the network? Evaluating the scalability of the network solution is critical. Network managers should attempt to assess its capability to expand -- in terms of size and performance. When performance requirements call for implementing ATM to the servers, or desktop, there is a need to also consider the potential of the solution to grow with the organization. Scalability can be evaluated in terms of: * The reliability of the server components: are they ready for large scale deployment? (See Factor 2) * Connection set-up performance: they vary among switch vendors * Transmission efficiencies, as defined by the Maximum Transfer Unit (MTU) * Support for multiple emulated LANs * Manageability Adaptec recommends a 9234 octet MTU for ATM hosts that need to communicate directly with other ATM hosts. As a contrast to the varied MTU sizes of LANE, CIP is only defined with a 9180 octet MTU size. [Image] Figure 3: Throughput The throughput of 155 Mb/s ATM at 1516 and 9Kbyte MTU sizes are compared above. These tests confirm that 9K tests result in superior throughput than 1516 octet tests because the host is forced to process nearly 6 times as many Protocol Data Units. Throughput levels are over 2 1/2 times as fast using the 9K MTU size. The overhead of the smaller MTU negatively affects host CPU utilization and effective throughput on the ATM interface. Figure 3 shows the effect of the lower software overhead associated with CIP. At both 155 Mb/s ATM and 25 Mb/s ATM, CIP throughput will outperform equivalent LANE tests. However, while LANE achieves only slightly lower throughput levels, it excels over CIP in many ways, including: multi-protocol support, broadcast support, ease of installation, and configuration options. Why support multiple emulated LANs? By supporting multiple emulated LANs, networks can deal with the fixed-MTU limitations of legacy technologies while also taking advantage of the unique efficiencies of ATM-to- ATM communication. MTU Size Allows in Octets Bridging to Comments ... Required for conversations between ATM and Ethernet clients. However, this low MTU size results in reduced efficiency for ATM-to-ATM 1516 IEEE 802.3 conversation and may be avoided in such cases. It Ethernet can be desirable to have ATM hosts in both a 1516 and a 9234 emulated LAN, one for talking to Ethernet hosts and the other for talking to ATM hosts. IEEE 802.5 4544 Token Ring 4 - Mb/s Default MTU for use over AAL5 as defined in RFC 9234 - 1626. This MTU size has been shown to result in peak throughput between ATM hosts. EEE 802.5 18190 Token Ring - 16 Mb/s Table E: Standard MTU sizes The ATM Forum has defined four MTU sizes for LANE. The intent of the different MTU sizes is two-fold: (1) to allow bridging between ATM and legacy networks, and (2) to offer various MTU options for performance purposes. Note that peak performance is achieved with the 9234 byte MTU. In one way or another, each of the LANE components must support multiple emulated LANs. Multiple copies of the LAN Emulation Client (LEC) can be used. The LAN Emulation Configuration Server (LECS) can point to multiple LAN Emulation Servers (LES) , and so on. When choosing LANE components, network managers should have a secondary backup solution to avoid a single point of failure. In the final stage of the evaluation process, the manageability of a switch manufacturer's product should be evaluated. Simple, graphical drag-and-drop interfaces as well as fault, and performance, monitoring capabilities are recommended. LANE Component Scalability Issues Should support multiple emulated LANs with any of multiple LEC MTU sizes. Peak efficiency can normally be achieved with an MTU of 9234 octets where ATM-to-ATM throughput is maximized and CPU overhead is minimized. Relatively low demands on the LECS except after a catastrophic failure such as a power outage. LECS is only LECS accessed at the initialization time of the LEC. Vendors should offer LECS redundancy. Should support all the standard MTU sizes, including 9234. Constantly directs connections for members of its emulated LES LAN by mapping MAC and ATM addresses. Vendors should offer LES redundancy. Should support all the standard MTU sizes, including 9234. Consider the amount of broadcast traffic that will be experienced in the new ATM emulated LAN. Consider the effects on the infrastructure of sending all broadcasts to the BUS, BUS and in turn, having the BUS resend the broadcasts to all nodes in the emulated LAN. Vendors should offer BUS redundancy. Should support all the standard MTU sizes, including 9234. Table F: Multiple Emulated LANs There are times when it's desirable to have a single-attached ATM host in multiple emulated LANs, such as when a node must talk to both Ethernet and ATM hosts. ATM implementations such as Adaptec's can support multiple emulated LANs on a single ATM interface card. --------------------------------------------------------------------------- Summary Before implementing an ATM network, network managers need to choose an ATM internetworking standard. Two standards-based alternatives exist to serve today's needs: LAN Emulation 1.0 (LANE) and Classical IP RFC-1577 (CIP). The advantage of LANE is that it offers extensive multi-protocol broadcast and multicast support by operating at the MAC sublayer. CIP can claim higher performance as a result of lower software overhead -- but strictly with IP. After choosing LANE or CIP, there are important factors to evaluate when choosing a switch vendor, including: interoperability, reliability and scalability. The information in this white paper was created to help provide an understanding of some of the issues related to the successful implementation of ATM technology for expanding the capabilities of your organization's network. --------------------------------------------------------------------------- Appendix: Typical ATM Implementations Example 1: Fatter Pipes Between Routers This application shows the use of ATM between routers to increase inter-router bandwidth. Note that since ATM is only being used for improved inter-router performance, LAN emulation techniques are not required (i.e. no server or desktop ATM connections). [Image] Figure 4: Fatter Pipes Between Routers Example 2: ATM Workgroup In this example, an ATM workgroup is connected to an existing network with an FDDI backbone. Most of the workgroup's traffic tends to be localized, which implies the clients talk to each other, or that the client's main server(s) is located within the workgroup. The connection to the main backbone allows ATM users to maintain corporate and Internet connectivity. Classical IP might be chosen over the LANE alternative (also pictured) when all the hosts are IP-based. A router is used to connect the workgroup to the rest of the legacy network. Note: An MTU size of 9180 is required by RFC-1577. [Image] Figure 5: ATM Workgroup (IP-based) [Image] Figure 6: ATM Workgroup Example 3: ATM Backbone The point of an ATM backbone is to deliver high performance ATM throughout the infrastructure. Legacy clients such as Ethernet and token ring are connected to the backbone via edge devices. Because of the heterogeneous protocols and the necessary bridging to legacy networks, LANE is the required standard. Note the position of the LEC, LECS, LES, and BUS. This the most intrusive example of an ATM implementation, because it requires replacing a significant portion of the existing infrastructure. The advantage of the backbone approach is it can deliver speeds of up to 622 Mb/s throughout the enterprise and will still be able to connect Ethernet and token ring nodes. [Image] Figure 7: ATM Backbone Example 4: Parallel ATM and Legacy Backbones The parallel ATM network is analogous to the workgroup environment, but larger in scale. In this application, either CIP or LANE may be implemented. Since the purpose of this parallel network is to eventually absorb the legacy network, LANE is the better choice because of its multi-protocol support. Note the primary and redundant LECS/LES/BUS. Assuming most traffic within the parallel network comes from within the parallel network, an MTU size of 9234 octets is desirable for peak performance. [Image] Figure 8: Parallel ATM and Legacy Backbone Example 5: Back-end ATM Network (Inter-server) One of the most straightforward implementations of ATM is a back-end ATM network tuned to boost the performance of inter-server connections. This environment benefits from off-loading existing infrastructure traffic and thereby increases the bandwidth available to other nodes in the network. In this example CIP is used because the main objective is to maximize inter-server throughput. Again, an MTU size of 9180 is used. [Image] Figure 9: Back End ATM Network (Inter-Server) --------------------------------------------------------------------------- Adaptec, Inc. 691 South Milpitas Boulevard Milpitas, California 95035 Adaptec Europe Belgium Tel: (32) 2-352-34-11 FAX: (32) 2-352-34-00 Adaptec Japan Tokyo Tel: (81-3)-5276-9882 FAX: (81-3)-5276-9884 Adaptec Singapore Tel: (65) 278-7300 FAX: (65) 273-0163 --------------------------------------------------------------------------- Literature: 1-800-934-2766 (USA and Canada) (510) 732-3829 Ordering Software: 1-800-442-7274 (USA and Canada) (408) 957-7274 Interactive FAX : (408) 957-7150 Adaptec USA Bulletin Board: (408) 945-7727 (up to 28,800 baud, using 8 bits, 1 stop bit, no parity) CompuServe: GO ADAPTEC Microsoft Network: GO ADAPTEC Internet ftp server: ftp.adaptec.com World-Wide Web: http://www.adaptec.com/ --------------------------------------------------------------------------- A Principal Member of the ATM Forum Copyright© 1996 Adaptec, Inc. All rights reserved. Adaptec, and the Adaptec logo are trademarks of Adaptec Inc., which may be registered in some jurisdictions. All other trademarks used are owned by their respective owners. Information supplied by Adaptec, Inc. is believed to be accurate and reliable at the time of printing, but Adaptec Inc. assumes no responsibility for any errors that may appear in this document. Adaptec, Inc. reserves the right, without notice, to make changes in product design or specifications. Information is subject to change without notice. Part Number: W920138-011 3/96