Fault tolerant Ethernet for Solaris

White paper by:-

Znyx Corporation

Editor's introduction:-

Network reliability has become as important to organizations today, as the availability of building services such as electricity and water. When these services stop working, the work environment is disrupted, and most people might as well go home. This article from ZNYX Corporation, describes how one manufacturer of Ethernet cards is supporting fault tolerance within the Solaris environment.

See also:-

Ethernet NICs, Rackmount SPARC systems, Articles, SPARC Resellers in the USA, SPARC manufacturers, STORAGE manufacturers, Rackmount SPARC Systems , Notebooks - SPARC , SBus & PCI cards, InfiniBand, iSCSI, SPARC news, STORAGE news

RAINlink for Solaris Feature Set

RAINlink for Solaris is an enhanced driver for ZNYX NetBlaster PCI and CompactPCI Fast Ethernet adapters. By embedding ZNYX' s innovative Redundant Array of Independent Netport (RAIN) technology, embedded RAINlink is a collection of enhanced network services that provides:

• High Availability (fast, automatic failover and failback)
• Scalability (trunking or link aggregation, dynamic load balancing)
• Flexibility (channels can be configured in a variety of modes)

High Availability

RAINlink offers multiple levels of high availability network integration:

• Multiple redundant ports on a single adapter
• Multiple redundant ports on multiple redundant adapters

Active standby

is the method used when either form of trunking is used. If one of the links in the trunk fails, traffic is routed to the remaining links. Available bandwidth is reduced but the trunk continues to function.

Inactive standby

is the method where a link (port) is configured to be a backup port but no traffic is routed to this link unless the primary link fails.

LINK FAILURES

Two types of events trigger a link failure:

• Link inactive due to hardware problems
• Link inactive due to absence of traffic

An inactive link due to hardware can be caused by:

• Cable damaged or disconnected
• Hub, switch, or adapter is powered down
• Malfunctioning board, port or network device in the connectivity path

An inactive link due to absence is implemented in RAINlink' s Watchdog Timeout feature. The Watchdog Timeout is a configurable parameter. The operator during configuration or installation of the NetBlaster and driver will specify a timeout interval. This interval may be in seconds, minutes or hours. If the RAINlink enabled driver does not see any valid incoming traffic on the port for the specified period of time, then the link becomes inactive and RAINlink switches traffic to a backup link. Valid Ethernet traffic must meet minimum packet size criteria and contain valid CRC.

The Watchdog Timeout feature covers a broad range of failures including defective NICs or any malfunctioning component in the connectivity path. This method works in either active or inactive standby modes. System administrators can setup ping scripts or other applications to create "keep-alive" test packets at desired intervals. Figure 2 shows how the Watchdog Timeout feature detects intermediate failures.

Scalability

RAINlink combines multiple Fast Ethernet ports into a single trunk and improves performance in both client-server (system-to-switch) and distributed computing or cluster (system-to-system) configurations. The trunk provides an Ethernet link in the no-man' s land between 100 Mbps and a gigabit link. The trunk appears as a single interface to the Solaris operating system, but actually functions by spreading traffic across multiple 100 Mbps links. The trunking method provides scaleable networking using existing CAT5 UTP cabling.

SYSTEM-TO-SWITCH TRUNKING

This method is used in large client-server networks. The clients are connected to a number of ports on a Fast EtherChannel-compatible switch. The server is connected using two or more ports to the switch. Figure 3 shows this topology. In this mode RAINlink balances the client traffic across the multiple links in the trunk. RAINlink also supports load balancing. If some clients are saturating a single link with too much traffic, RAINlink will spread this traffic across all the links in the trunk.

SYSTEM-TO-SYSTEM TRUNKING

This method is used in applications where servers or systems need to transfer large amounts of data. This could be cluster applications or other applications that require more than 100 Mbps bandwidth to transfer data. Systems are connected in a port-to-port, full-duplex mode. No hubs or switches are used. Operating in this mode provides collision free, full-duplex operation (100 Mbps each direction on each port), with minimal interpacket gap. Figure 4 shows this topology.

SPARC(R) is a registered trademark of SPARC International, Inc. SPARC PRODUCT DIRECTORY(SM) is a service mark of SPARC International, Inc used under license by ACSL. Products using the SPARC trademarks are based on an architecture developed by Sun Microsystems, Inc.