A bend, don't break strategy for SPARC servers Resilience combines three SPARCstation 5s in one server By Rebecca Gold

November  1996

Mail this article to a friend

Palo Alto -- On October 11, two curious rats scampered down a cable conduit at Stanford University's independent power plant, igniting a fire in vital switching gear. The rodents knocked out electrical power on Stanford's campus, and stranded tens of thousands of San Francisco-area Internet users as a result. BBN Planet, an Internet service provider to many large firms in the area, houses its local point-of-presence at Stanford.

It took nearly 24 hours to restore full Internet service to such high-profile customers as Apple Computer, Sun Microsystems, PC World magazine, Computerworld, JavaWorld, as well as almost 400 other companies.

The disruption was a harsh reminder that while the Internet is a relatively new technology, it is a service on which millions depend. If the mighty Internet, which was designed to withstand nuclear attack, is vulnerable to two rats, can we consider our common servers safe from more routine maladies?

Resilience Corp. is a Palo Alto, CA start-up that hopes Stanford's rats will remind Unix server buyers that bad things happen to good hardware, and that fault-tolerant systems might be a smart idea for many organizations. Resilience Corp. is the culmination of years of SPARC technology developed by several companies. Resilience founder and president Mark Johnston started at Axil Computer, a SPARCstation compatible maker. He then founded Cycle Computer, where he developed the Cycle 5 motherboard swap for users of old, feeble SPARCstations.

Fault-tolerant hardware is de rigueur in pockets of the computer world today, including airline ticketing, telecommunications, and financial houses. The fault-tolerant concept is simple: let several, intertwined systems run the same applications simultaneously. If one CPU, hard disk, or power supply fails, the remaining subsystems soldier on without interruption. Unlike its similar technologies fault resilience and high availability, fault tolerance provides constant service. But fault-tolerant hardware is proprietary and expensive, and as a consequence, it's relatively rare in common server rooms.

Sun all over again
Sun Microsystems was formed on the notion that one could take off-the-shelf components and build a cheap but powerful workstation. Resilience is hoping to find success with the same idea in fault-tolerant servers. In a nutshell, Resilience packages three SPARCstation 5-class motherboards in a cabinet with several power supplies, and a disk. The motherboards are home to a set of patented ASICs that allows the motherboards to communicate with one another. Collectively, the ASICs are known as the Resilient Bridge -- the key both to the successful failover of the system and to Resilience Corp.'s raison d'etre.

Dubbed the Resilient 110, the system marks a new low for fault-tolerant servers: $21,995, or about one quarter the price of similar systems. The price includes the three 110-MHz SPARCstation motherboards, each one hot-swappable and containing 32 megabytes of memory. The Resilient 110 runs all Solaris applications, and is available in both desktop and rackmount configurations.

All three motherboards run the same applications at all times in lock-step. Each checks the other two constantly for signs of life via the switch. When one fails, the others censure it, shut it down, and carry on. All three computers come with their own subcomponents such as I/O controller, memory, and power supply. The only things they share are information with each other, and the ability to detect when something is amiss. The modules are hot-swappable, allowing the operator to fix the failed CPU board or subsystem in minutes, without having to stop the application.

According to Ken Schmitt, marketing communications representative at Resilience, the Resilient 110 was built around the idea that not everybody needs the "heavy-duty power of an UltraSPARC," and that by using the SPARCstation 5 architecture, Resilience was able to offer something that's affordable to users who want reliable e-mail and Internet servers, for example, without messing with software-based near-real-time solutions.

The Resilient 110 system was released in June and is available now through value-added resellers. -- by Rebecca Gold

Resources

• Resilience Corp.
  http://www.resilience.com/
• Cycle Computer
  http://www.cyclecc.com/
• Network Appliance
  http://www.netapp.com/
• Invincible Technologies
  http://www.invincible.com/
• Auspex
  http://www.auspex.com/
• "Invincible servers find a niche"
  /sunworldonline/swol-11-1996/swol-11-invincible.html

If you have technical problems with this magazine, contact webmaster@sunworld.com

URL: http://www.sunworld.com/swol-11-1996/swol-11-resilience.html
Last modified: