Highly Available Networks

ManualsBrandsHP ManualsSoftwareHP Serviceguard Software

Pamela Williams Dickerman

Advanced Technology Consultant

Michael Hayward

Hewlett-Packard Company

Table of Contents

Abstract

Single Points of Failure in a Local Area Network

Bus Topologies

Star Topologies

Ring Topologies

Eliminating Single Points of Failure in a Local Area Network

Client Recovery

Multiple subnets and gateway protection

Local Failover Functionality of MC/ServiceGuard

Eliminating SPOFs in LAN Topologies

Highly Available Bus Topologies

Highly Available Star Topologies

Highly Available Ring Topologies

Further Considerations for Highly Available Networks

Abstract

Networking has many hardware components; each could be a Single Point of Failure (SPOF). The key

to eliminating failures within the network are understanding the topologies being used, understanding

the failure points within those topologies, and removing these failure points from the network.

There are hardware and software products such as dual attached FDDI cards and HP’s

MC/ServiceGuard which provide increased network availability.

What type of hardware redundancy is required to protect against SPOFs?

What type of software is needed to work with the hardware?

How do these products fit together?

This paper discusses how SPOFs in the network can be eliminated or how the fault can be segmented to

such an extent that only a small percentage of users will be affected during an outage.

Summary of content (11 pages)

PAGE 1
Highly Available Networks Pamela Williams Dickerman Advanced Technology Consultant Michael Hayward Hewlett-Packard Company Copyright 1996 Hewlett-Packard Co., Inc.
PAGE 2
Single Points of Failure in Network Topologies The key to eliminating communication downtime is to identify and eliminate the SPOFs within the network. There are many different types of network topologies, each with their own SPOFs. This paper will discuss three main topology types: Bus Star Ring Bus Topologies A bus type network is one where each system taps directly into the same cable, has a matched impedance, has only two ends, and all signals are terminated at the cable ends.
PAGE 3
Figure 2: Star Topology: Using Twisted-Pair With Twisted-Pair, the SPOF is changed from the cable to the hub. If a single segment from a system to the hub or if the LAN card on that system fails, only that system is unable to use the network. Optical fiber based networks use a concentrator, or a switch, rather than a hub. The two most common protocols running over optical fiber are Fiber Data Distributed Interchange (FDDI) and Fibre Channel (FC) with FDDI currently the largest percentage.
PAGE 4
Figure 4: Ring Topology: DAS FDDI without concentrators The two most common ring topologies are Token Ring and DAS FDDI. This paper will focus on DAS FDDI. Unlike the SAS configuration, there is no need for a concentrator except to connect to SAS based systems. In addition, DAS FDDI has a self-healing capability. If there is a segment failure, DAS can wrap the failure and send the data in the other direction. Therefore, as shown in Figure 4, there is no SPOF that would bring down the entire LAN.
PAGE 5
Figure 5: Separate Subnet Example In Figure 5, both systems have been configured with two LAN controller cards and with two separate cables between Server1 and Client1 to create two subnets. Each LAN card has a different IP address. If the cable connecting to lan1 fails, this would bring down access to the server’s IP address 15.0.0.1. To recover, the client could connect to Server1 via the 16.0.0.1 address through lan2. The client which normally does a telnet 15.0.0.1 would simply do a telnet 16.0.0.
PAGE 6
Server and client are required to have MC/ServiceGuard installed to provide local switch functionality if both on same LAN. Requires two HUBs cross connected for no SPOF Transparent to Applications Fast Failover Requires idle LAN host adapter As will be discussed in the next section on LAN Topologies, the MC/ServiceGuard approach uses two hubs connected together to create one LAN with alternate paths. After a failure, the 15.0.0.1 address is transparently moved to the lan2 controller card.
PAGE 7
Figure 8: Highly Available Bus Topology To eliminate the bus and interface cards as SPOFs, redundant busses are used. These two separate busses are linked together by a single hub. Since clients are usually a PC or workstation and MC/ServiceGuard does not run on these types of systems, we must use a segmentation scheme to isolate faults to the smallest possible area of impact.
PAGE 8
Figure 10 shows you how to isolate the clients on separate subnets via routers. Like the previous example, the local subnet for the servers is made highly available by using two cable segments and a hub. However, connection to the servers is made via the routers. The clients have a single point of failure at the router. If a router fails, the clients connected via that router will be unable to communicate with the server. The client’s connections through the other router will still be up and communicating.
PAGE 9
as duplicate IP addresses are avoided. For customers who are not concerned about maximizing network throughput and do not want to take on the task of configuring gated on all their systems, the default route set to local will work as an alternative. Configure each local node as the default router and it will broadcast all IP packages destined for the other subnets.
PAGE 10
In Figure 13, routers are added into the star topology. The same rules apply for this configuration as discussed in "Figure 11: Highly Available Bus Topology with Cross-Connected Routers". The cables can be replaced with token ring or optical fiber and the hubs replaced with concentrators. Everything would work the same as discussed for the twisted-pair bus topology.
PAGE 11
Further Considerations of Highly Available Networks Eliminating the Single Points of Failure in a network and having client recovery mechanisms in place are only the first steps in creating highly available networks. Ongoing planning, clear documentation of the current networking environment, and written procedures for replacement of failed parts must be well understood. It takes careful planning and investigation, but highly available networks can be setup and maintained.