IBM is on track to release the next version of its Unix-based AIX operating system in the second half of this year.
That release will include many new features and technologies currently unavailable to users, and will come more than a year after the Armonk, N.Y., company released the current AIX 5L version, 5.1.
“This will be a significant functional addition release, and youll see technologies brought forward that are not available today. There will be significant feature add-ons,” Mike Harrell, an AIX product director, told eWEEK in an interview on Tuesday, declining to comment further.
The AIX operating system will continue to run only on IBMs own eServer Power product family. “There will not be an Intel version of it. While we produced a version of AIX that ran well on the first-generation [Intel] Itanium platform, we no longer build an AIX version for Itanium. We have withdrawn from that program and have no plans to rejoin,” Jim McGaughan, director of eServer marketing, told eWEEK.
IBM will also announce on Wednesday three new high-performance features that have been added to the current AIX 5.1.
The new features–technical large page, scheduling and memory affinity–are designed to “supercharge compute-intensive applications” and improve the speed of bandwidth-intensive workloads, like business intelligence applications that search massive corporate data warehouses, as well as high-performance computing applications, IBM executives said.
“These three elements work together to provide a key function that our performance customers will find beneficial and allow them to take advantage of functionality available in the Power 4 architecture,” Harrell said. “The new technologies have been incorporated in the latest maintenance release of AIX that is now available to customers.”
The technical large page feature allows AIX to quickly page large amounts of memory back and forth between local memory, the Power 4 system and the storage device in chunks of 16MB.
Scheduling affinity then takes this high-performance workload and locks it internally to the server to a subset of processors and system memory, where memory affinity takes the technical large pages and shifts them back and forth between the servers memory and the processors running the workload.
“So this is a triangle of technologies all working together to utilize and harness the architecture running this workload and give it a supercharge and a whole lot more throughput than it might have otherwise,” Harrell said.
While the three latest technologies are geared toward high-performance technical and scientific computing applications, the business intelligence community will also benefit from the use of this feature combination “because databases tend to be very large and they have to be queried very quickly, with a lot of information being moved back and forth,” he said.
Jonathan Eunice, president of research group Illuminata Inc. in Nashua, N.H., said he expects primary demand for this new technology to come from the technical computing space.
But the page size and scheduling affinity could be a utility for very memory- or resource-intense computing, including databases and specific large codes, “but thats not the No. 1 group thats requiring it,” he said.
Affinity functionality makes the most difference on the largest configurations, and as processor counts increase, so does the need and benefit of the optimization, he added.
IBM will also announce Wednesday that it is working on an AIX toolbox for grid applications, based on the open-source protocols from Globus, which will be available this summer.
“This is a second-generation tool kit for developers to help them build applications and middleware intended to drive the new grid computing initiative that allows users to share supercomputing power, data and applications as easily as information is shared over the Web,” Harrell said.
The toolbox will be offered for free, and IBM will also offer service and support for it. Among the new features in the toolbox are a Message Passing Interface, Advanced Reservation and IBM LoadLeveler integration–a batch job scheduling facility that matches job requirements with available resources.
Illuminatas Eunice said while IBM has been at the forefront of defining grid protocols and was instrumental in the merger of Web services and grid protocols, these were open protocols that anyone could implement.
“The question is not whether you can do grid computing on the various Unix platforms, but who makes it the easiest and most seamless. And bundling recent versions of the protocols sure helps in that regard,” he said.
