SGI's Altix Ready to Rock with Madison Itaniums.By Timothy Prickett Morgan Having launched itself into the Itanium server market with the Altix 3000 machine in January, SGI (SGI, Sunnyvale, CA, www.sgi.com) A manufacturer of workstations and servers, founded in 1982 by Jim Clark. The company was founded as Silicon Graphics, Inc., but changed to its acronym in 1999. Inc is thrilled that Intel Corp has finally announced the "Madison" versions of its 64-bit processors. To SGI's way of thinking, the impending im·pend intr.v. im·pend·ed, im·pend·ing, im·pends 1. To be about to occur: Her retirement is impending. 2. announcement of the 1.5GHz Madison kickers to the second generation 1GHz "McKinley" Itanium 2 processors was in part stalling the market for big Linux servers in the high performance computing market. Any time a new processor is coming to a market, it tends to put a damper damp·er n. 1. One that deadens, restrains, or depresses: Rain put a damper on our picnic plans. 2. An adjustable plate, as in the flue of a furnace or stove, for controlling the draft. on sales of the existing machine, even if, in the case of the McKinley and Madison Itanium 2 processors, the chips are plug and software compatible. It's a psychological phenomenon as well as an economic issue. Madison processors offer about 50% more oomph than McKinleys, and while we didn't know it until this week, they are being sold at the same price. At several grand a pop for each Itanium processor, only a company that really needed an Altix 3000 server and its unique shared memory (1) Using part of main memory to support a low-cost display circuit that does not have its own memory. See shared video memory. (2) The common memory in a symmetric multiprocessing system that is available to all CPUs. See SMP. 1. space for supporting very large Linux workloads would have ponied up the cash until Madison was here. Not that SGI seems to need Madison. The fact that six Altix machines are already on the Top 500 list of supercomputers after only being available for twelve weeks is a good sign that Linux on Itanium is getting some traction in the HPC (Handheld PC) A palmtop computer that weighs less than one pound and runs specialized versions of popular applications. Microsoft coined the term for its Windows CE operating system, which is an abbreviated version of Windows. See Pocket PC. space. Andy Fenselau, product line director for the Altix 3000 machines at SGI, says that there was an order of magnitude A change in quantity or volume as measured by the decimal point. For example, from tens to hundreds is one order of magnitude. Tens to thousands is two orders of magnitude; tens to millions is three orders of magnitude, etc. higher interest in the media when it came to the Altix line than it expected, which has helped spread the word about the Altix line. But sales are ultimately what count. In the first twelve weeks of sales, SGI had 63 customers who have acquired Altix machines. Nine of the customers were acquired by government agencies, 28 were life sciences research institutions, another 17 were physical science research institutions, and the remaining nine customers were in various industries such as oil and gas exploration. Not coincidentally, these sectors have the most modern code and have the most ISV (Independent Software Vendor) A person or company that develops software. It implies an organization that specializes in software only and is not part of a computer systems or hardware manufacturer. support for both Linux and Itanium. SGI did not, for instance, expect much activity in the manufacturing industry. "This is the most ISV-driven part of the HPC market," he explains. "With any new architecture, it takes time. But we are beginning to see results here, now, too." The Altix 3000 machines are not the same type of Linux clusters that IBM (International Business Machines Corporation, Armonk, NY, www.ibm.com) The world's largest computer company. IBM's product lines include the S/390 mainframes (zSeries), AS/400 midrange business systems (iSeries), RS/6000 workstations and servers (pSeries), Intel-based servers (xSeries) Corp, Dell Inc, and Hewlett Packard Co have been selling and, as the Madison Itanium 2s rolled out, IBM and Dell spent most of their time talking about. The Altix 3000s borrow the NUMAflex clustering and shared memory technologies that SGI developed for its Origin 3000 Irix machines. Under the NUMAflex cache-coherent, non-uniform memory architecture, each processor runs its own instance of an operating system operating system (OS) Software that controls the operation of a computer, directs the input and output of data, keeps track of files, and controls the processing of computer programs. , just like a parallel cluster does. Most servers today are either uniprocessor machines or machines that use symmetric multiprocessing See SMP. (parallel) symmetric multiprocessing - (SMP) Two or more similar processors connected via a high-bandwidth link and managed by one operating system, where each processor has equal access to I/O devices. (SMP (Symmetric MultiProcessing) A multiprocessing architecture in which multiple CPUs, residing in one cabinet, share the same memory. SMP systems provide scalability. As business increases, additional CPUs can be added to absorb the increased transaction volume. ) technology to create a single virtual processor from many processors that all share a single main memory address space. NUMAflex does this, too. All of the processors in a NUMAFlex cluster can access the same shared pool of memory. Applications run in memory, and any time processors have a single memory space to play with, it simplifies programming and allows bigger programs and more programs to run more efficiently on a given machine than is possible in a loosely coupled See loose coupling. machine like a Beowulf Linux cluster or a massive parallel Unix cluster. Another benefit of an Altix 3000 is that memory capacity in an SGI NUMA (Non-Uniform Memory Access) A multiprocessing architecture in which memory is separated into close and distant banks. NUMA is similar to SMP, in which multiple CPUs share a single memory. However, in SMP, all CPUs access a common memory at the same speed. cluster can scale independently of processor count. This means that companies can buy the minimum amount of CPUs necessary to support their workloads and buy the maximum amount of main memory to goose the performance of their workloads. The two are not tied, as they are in more traditional Linux clusters. If you want to add more memory in a traditional Lintel cluster, you have to add more servers, and if you want to add more processors, you have to add more memory. The Altix can support up to 4TB of shared main memory. Fenselau says that in retrospect SGI was a little late in the McKinley cycle - about six months - but that SGI will be right on time with Madison, ready to ship products this week if that is what customers want. The company has sold a number of departmental machines that span from 4 to 16 processors and has sold a bunch of machines with 128 to 256 processors. (The biggest node size that the Altix supports is 64 processors in a single node.) The largest Altix machine sold to date has 416 McKinley processors. Oddly enough, this machine does not appear to be on the Top 500 list. The typical Altix machine is configured with 3GB to 4GB of memory per processor, on average. A typical SMP machine has 1GB to 2GB of main memory per processor in a Lintel cluster. As Intel rolls out future Itanium processors, they will plug into the existing Altix machines. Fenselau says that they Altix servers were designed from the get-go to support many Itanium generations without requiring an upgrade of the NUMAflex technology. That said, SGI will roll out the next generation of its NUMAflex clustering in the second half of 2004, presumably pre·sum·a·ble adj. That can be presumed or taken for granted; reasonable as a supposition: presumable causes of the disaster. for both Origin-MIPS-Irix and Alrix-Itanium-Linux machines. |
|
||||||||||||||||

Printer friendly
Cite/link
Email
Feedback
Reader Opinion