On March 20, IBM continued its push into the Windows 2000 (Win2K) enterprise server space by announcing two new 4-way servers, the Netfinity 7100 ($7000) and Netfinity 7600 ($9940). Netfinity is IBM’s departmental server offering based on Intel architecture.
In an interview with Tom Bradicich, director of Architecture & Design for IBM Netfinity Servers, he described how IBM is hardening its Netfinity servers for use with Win2K. IBM’s efforts span the range of activities from improving software quality, adding new features to hardware, and improving its servers’ fault-tolerant capabilities. According to Bradicich, "Windows 2000 couldn’t have come at a better time because it adds capabilities that allow us to do this kind of work and be successful." Much of IBM's work leverages its X-architecture development from previous OSs using techniques the company developed in the past on the Win2K platform.
IBM has added an Advanced System Management Processor to its Netfinity systems that monitors the health of the system and can control the server remotely. At the moment, IBM is using its Remote Connect feature to offer service with a 99.99 percent uptime guarantee. That commitment translates into roughly 1 hour of downtime per year.
The X-architecture provides Netfinity systems with a fail-over node. This fault tolerance has let IBM experiment with methods to make its systems more reliable. For example, IBM offers a software regeneration technology that moves activity to an inactive mode and then reboots or refreshes the other nodes, one at a time. Netfinity also offers ActivePCI, which lets you incrementally add more networking capability to the system while it's up and running. Another new feature, a power fuel gauge, checks the electricity delivered to the server to predict power requirements. IBM developed this last technology as part of its OnForever initiative.
Bradicich said that IBM is attacking reliability from all angles. New technologies you can expect to see in the company's servers include:
• Chipkill memory, special memory that lets a server recover from memory chip failure, a feature IBM is offering for 8 bits of memory failure
• Special diagnostic features for IBM's device drivers
• Hot swap memory
• Extensive scalability As part of the Netfinity announcement, IBM has said it plans to mate its two new Netfinity servers with its Storage Area Network (SAN) offerings. These types of announcements are part of a growing trend of for enterprise server vendors to differentiate their server products by selling server and storage combinations. IBM’s announcement is unique in that the company's new line of SANs devices are full fibre channel implementations. Only EMC’s CLARiiON division offers full fibre channel units in the industry. IBM’s FastT500 RAID controller unit is fast, offering a 300MBps throughput. Typically, controllers are in the 100MBps range. This controller can manage 220 hard disks or up to 8TB of storage in a set of connected devices, and costs $36,750. A new fibre network adapter card, the FastT Host Adapter Card ($1500), is now available that wraps IP packets inside fibre channel frames for a 300Mbps throughput. Additionally, IBM has added the Netfinity Fibre Channel Storage Manager Partition Enhancement 1.0 ($7400) to its Netfinity Fibre Channel Storage Manager software offerings. This partitioning software manages disk space on storage devices between servers, applications, and users. The partition manager can offer dedicated partitions on a storage device for up to eight different hosts. IBM is also working with Intel, Microsoft, and PC Doctor to develop a common diagnostic model, and is active in the Distributed Management Task Force. IBM will have an announcement about its next-generation Summit architecture soon; that architecture will appear in 2001. Bradicich also mentioned efforts underway in his group to provide a software predictive failure analysis capability in the IBM Netfinity server line. This tool analyzes the patterns of counters and indicators before failures and learns to recognize impending failures and intervene before they occur.