The Return of the Mainframe?
投稿者: Arthur Cole 2009/12/10 13:55:56
Server consolidation is the name of the game these days, which means enterprise architectures will probably start shying away from large numbers of small machines in favor of more powerful devices capable of increased VM counts.
But will this trend extend all the way back to the mainframe?
IBM seems to think so, coming out with a renewed focus on big iron that the company says can deliver virtual and cloud environments at a fraction of the cost of commodity x86 servers. The new systems in IBM’s line certainly don’t measure up to the massive mainframes that were once the mainstay of the IT industry, but they are nonetheless the most powerful systems on the market.
The latest entrants are a pair of new Enterprise Linux Servers, which start out at a cool $212,000 for a two-processor model. The company has combined either the Novell SuSE or Red Hat OS with its own z/Virtual Machine system in what it says is a lower-cost, integrated stack, with the idea that it will prove competitive against large-scale multicore clusters that are rapidly becoming the go-to solution for highly virtualized environments. Among the chief advantages is the ability to add capacity without pulling existing environments offline.
The new systems come on the heels of a dramatic upgrade of the System z software portfolio. The package includes a new SQL-compatible data management stack, a new Tivoli handling and reporting system, upgraded NetView and Asset Discovery tools, and a pair of new development tools linked by a common repository. The goal here is to convince users that mainframes can provide the same resources as distributed architectures, without the logistical and management hassles.
At the same time, the company is dialing back on some of its more innovative blade server technologies. Chief among them is the QSZ2, powered by the PowerXCell, a hybrid processor consisting of a pair of 64-bit Power cores and 32 of the company’s “synthetic processing elements” (SPEs). The Cell technology is still slated for future generations of supercomputers and the upcoming Power7 line. But low-end devices intended for massive cluster environments seem to be off the radar for now.
IBM is not the only one touting mainframes at the moment. Fujitsu recently added a pair of machines to its BS2000/OSD line using 65 nm processors that both improve system capacity and lower costs some 20 percent. The S210 unit scales up to 16 cores with 256 GB of memory and touts eight I/O processors capable of providing up to 256 peripheral channels, plus another 64 Fibre Channel interfaces. The mid-range S175 supports up to three processors with 64 GB of main memory.
Clearly, mainframe solutions are optimal in specialized, large-scale environments where resource requirements are fairly well-known ahead of time. While it seems that some of the scalability issues of the earlier machines have been addressed, rapidly changing environments that see constantly shifting data loads would probably do better with a distributed architecture.
Still, IBM and Fujitsu, at least, must feel there are enough staid enterprises out there to justify continued investment in big iron. And with the rise of new cloud service providers with their massively dense data centers, they just may be right.