128-Way NT Coming Your Way
Times N Systems, a privately held start-up company in Austin, Texas, is promising 128-way Windows NT clustering technology, with 20 to 30 percent faster performance than traditional clustering technology for clusters of similar size.
May 30, 2000
Times N Systems, a privately held start-up company in Austin, Texas, is promising 128-way Windows NT clustering technology, with 20 to 30 percent faster performance than traditional clustering technology for clusters of similar size. Times N Systems showed a working prototype at a Data Warehousing conference in Dallas during the week of May 8 through 12. Theodore Scardamalia and Lynn West, the former manager and senior architect, respectively, of IBM's Advanced Technology Development Group, started this company after leaving IBM in 1994. Scardamalia and West spent years developing the Processor Teaming architecture, a technology quite different from SMP, NUMA, and clusters. At the same time, IBM had a second research team developing a different solution for better clustering. The company had no plans to integrate the two research teams and went with the other team's work, picking up Power Parallel technology and releasing the Processor Teaming research to the public. The theory behind Processor Teaming is that conventional clustering has problems with memory contention, data transfer bottlenecks, and latency. The Processor Teaming solution used a "share what you can" approach, making the individual members of a cluster much more independent. Although the details are fuzzy, Times N Systems promises the symmetric scaling of SMP with the scalability and I/O performance of massively parallel processing (MPP). The technology gives each processor its own OS—the first target is NT; Linux will follow—making the processor more independent than in traditional clustering technology. Times N Systems' technology coordinates each of these different OSs through the OS's native APIs. The technology also gives each processor its own memory and attaches a separate memory pool to the cluster.
About the Author
You May Also Like