Network essentials - Migration made simple

1 Factors influencing the decision to migrate

eadaches for network managers, but these tips should help ease some of the pain. - Token Ring technology is complex, and there is an increasing lack of trained support personnel.

- Ethernet and Fast Ethernet equipment prices are much lower than those for Token Ring.

- Ethernet provides additional future-proof scalability, as existing and emerging high-speed technologies seamlessly integrate with Ethernet, while Token Ring support is questionable in many cases.

- There is limited long-term industry commitment to high-speed Token Ring.

2 Look at your resources, move at your own pace

First of all you must remember that no migration occurs overnight. Depending upon your financial and human resources, the migration could take months or even years. During the migration, Token Ring and Ethernet devices still need to communicate with each other. Therefore, you need a migration strategy that allows you to move desktop users or servers to Ethernet at a reasonable pace, while maintaining communications with the Token Ring devices.

3 Minimise disruption

It may be prudent to keep your existing users on Token Ring and install Ethernet in new locations or in new desktops and servers, or even when a new application is rolled out. This gradual approach reduces the amount of disruption for existing computers and devices. If required, Token Ring switching can be used to provide increased performance for the Token Ring infrastructure, while Ethernet switching is installed for the new devices.

4 Replace backbone bridges with Token Ring switches

As a Token Ring network becomes saturated with traffic, IS managers can replace multiport backbone bridges in the data centre with Token Ring switches to boost capacity where needed. Token Ring switching transparently replaces existing bridging solutions in addition to simplifying adds, moves, and changes. Each port on the switch represents its own ring, which can communicate with other switched rings without external bridging or routing. Essentially, each port becomes an internetworked Lan.

5 Switch your Token Rings

The main drivers for ring segmentation are power users with local servers.

By subdividing the local rings, with power users and their local servers on the same segment, you can increase the available bandwidth for the power users while offloading that traffic from the rest of the network.

Additionally, servers used by a broader range of end stations are directly attached to switch ports, gaining access to the full Token Ring bandwidth.

The resulting performance improvement for all users is achieved at the cost of a few ports in the backbone switch.

6 Connect workgroups to the Ethernet backbone

Adding a Token Ring switch at the floor level, for example, improves performance on the local ring by removing server traffic, and it hands the server 100-200Mbps of dedicated bandwidth. The same switch can also split 50- to 80-user rings into much smaller rings, each with fewer users. A single uplink to the high-speed backbone provides all workgroup rings with high throughput access to mainframes and server farms.

This solution produces three key benefits:

- It reduces the impact of heavy users on other users.

- Each user enjoys improved performance while the existing network workgroup infrastructure is preserved.

- Network managers can segregate older 4Mbps Token Ring end-stations from 16Mbps Token Ring end-stations. With a simple reconfiguration, users with 16Mbps capability get a four to one bandwidth increase, while investments in 4Mbps adaptors remain intact.

7 Limitations of Token Ring switches

There are some important capability limitations to consider when implementing Token Ring switches, however. Some Token Ring switches support source-route bridging only and will not pass transparent packets, but ideally you want switches that support both transparent bridging and source-routed frames.

8 Move to higher bandwidth

Because it merely hastens the overload on the 16Mbps Token Ring backbone, Token Ring switching can only be considered a preliminary step to full Ethernet migration. But since Token Ring technology is not going to disappear overnight, IT managers must find a higher-bandwidth alternative for the network core.

9 Replace Token Ring backbone

As user populations and bandwidth requirements increase, the campus/building backbone in the data centre becomes the obvious candidate for upgrading to a high-speed technology. Currently, there is a choice of three available to replace existing Token Ring backbones: ATM or Ethernet, either in Gigabit or 100Mbps form.

Each technology has its advantages and disadvantages. While ATM might be the right choice for some customers, for pragmatic reasons the market trend today is migration to Fast Ethernet and Gigabit Ethernet technologies.

It is important to remember, however, that it's not a straight choice between Ethernet and ATM.

10 Dual backbone difficulties

A network with dual backbones and multiple Token Ring interface coupler (TIC) attachments, configured with identical MAC addresses will not function as expected unless a source-route bridge sits between the TIC attachment and the rest of the network. By using a true source-route transparent (SRT) switch that can forward both source-route and transparent packets, you can improve throughput while maintaining compatibility with the current environment.

11 ATM servers as a migration aid

Dual-homed ATM servers - servers installed with both ATM and Token Ring server network interface cards (NICs) - allow users to share resources without the need for translational bridging. Using ATM as a transport, both Token Ring and Ethernet data travel on the same fibre cable. There are no compatibility issues with this solution because Token Ring and Ethernet users reside on separate emulated Lans (Elans) - the ATM servers can be members of both Elans.

ATM, however, does not support source-route-transparent bridging, so direct Ethernet workstation to Token Ring workstation communication (or 'translation') is not available. An ATM router is required for translation between Token Ring and Ethernet Elans.

12 Replace end-station NICs

The final migration step is to replace NICs in Token Ring end-stations.

Because of this gradual approach, the cabling plant can also be converted in parallel, allowing each customer to select the most appropriate migration path for replacing NICs on as many workstations as needed.

13 Easy does it on the desktop

Many companies are considering moving from IPX, SNA, and NetBIOS protocols to a pure IP client as part of their migration to Ethernet. This approach, although it takes more effort, can be done at the same time that the desktop's NIC is changed to Ethernet. A pure IP desktop eliminates the need for multiprotocol routing and translational bridging support between the Token Ring and Ethernet networks.

14 Data Link Switching (DLSw)

Multiprotocol routers with DLSw functionality have interconnected dissimilar protocols over the Wan for years. The encapsulation technique was originally introduced by IBM, and in March 1993, it was adopted by the Internet Engineering Task Force (IETF) as RFC 1434, and enhanced and superseded by RFC 1795 in April 1995.

DLSw Version 2 (RFC 2166) was developed in 1997 to address scalability and improve connection set-up time. Among other major benefits, DLSw v2.0 reduces Wan backbone traffic and TCP overheads, and increases the number of supported connections.

15 Migrate enterprise Wan networks using DLSw

Because of the timing sensitivity of SNA sessions, DLSw is the only dependable technique for transporting SNA over IP networks. DLSw routers reduce the likelihood of timeouts by providing local polling and logical local termination of the data link - often called 'spoofing'. By keeping the acknowledgment local, routers also reduce the amount of traffic traversing the wide area link. Additional DLSw advantages include the ability to route around link failure, support for an extended source-route bridging hop count, and NetBIOS name caching, which greatly reduces NetBIOS broadcasts.