IBM next week will debut its next generation X-architecture to boost the fortunes of its Intel-based servers in the cloud computing and big data markets, according to sources familiar with the company’s plans.
The X6 Architecture, which further evolves the X5 architecture’s concept of disaggregating processors from memory in large racks of servers, is a complete redesign of a traditional server rack with significantly increased density and server performance. X6 is built around a “mid-plane architecture” that also borrows from Intel Corp.’s QuickPath Internconnect (QPI) technology to enable fast memory-to-memory transfer rates, sources said.
The renewed emphasis on Intel servers follows speculation last year that IBM was shopping its Intel-based xSeries of servers around, largely because profits on the line were falling and the technology was becoming too commoditized for IBM’s liking.
The upcoming X6 employs the concept of “books,” or modules, which contain an Intel processor and/or dual in-line memory modules (DIMMs). There will be a Compute Book with a CPU and up to 24 DIMMs and a Storage Book that can be stuffed with up to 6.4 terabytes of xFlash storage with up to 12.8 terabytes of Serial Attached SCSI (SAS), according to sources.
“This isn’t trivial stuff to pull off. This (architecture) allows storage-to-memory and memory-to-memory to work at about the same speed. You should effectively be able to do almost any compute or memory intensive task completely in memory,” according to an East-coast-based consultant familiar with IBM's X6 strategy, who requested anonymity.
One of the potential benefits of the new architecture is allowing corporate users, particularly those in the financial services and pharmaceutical industries, to make more effective use of analytics and large databases at the departmental level in terms of productivity and cost savings.
“One of the complaints people have about all these I/O hungry applications is how to push analytics and large databases down so departments can work with them. But with (IBM’s) use of flash memory, it looks like they can do that, and save (users) a bunch of money,” the consultant said.
“You can now take your existing storage and put that in the background like a second tier of storage because you aren’t working with spindles or rotating drives anymore,” said one IBM insider who requested anonymity.
IBM next week will claim that the new architectural improvements should result, on average, in savings of 43% with three times lower latency for many IT shops, sources said.
“You can directly access three times the amount of memory through this memory channel storage, which is basically flash memory,” according to the East coast IT consultant. “So you will get very low latency because of the way they have integrated caching technology with the memory bus.”
Some observers believe the new X6 architecture could give the series a hipper technology image, thereby attracting more interest among potential buyers.
“The xSeries is still a viable business but it does need a shot in the arm for sure. I think they see this announcement as doing just that,” said one industry observer who requested anonymity.
X6 features include higher resiliency
The new architecture also introduces a higher degree of resiliency, including an improved ability to intelligently identify and work around failures and admins to take preventative steps to correct them. IT professionals can carry out a range of different system management tasks using virtual tools, which should help keep IT support costs down, according to sources.
QPI is an architecture first put into chips by Intel in 2008. It has also been used in IBM xSeries servers before. It allows for memory-to-memory connections between chips within a server to boost performance. In the case of X6, this technology is also being used to connect disks, both Flash and Serial Attached SCSI (SAS), to the CPU at the speed of memory.
QuickPath Interconnect also allows CPUs to access memory in a system other than that which is directly attached to the chip, using what’s called a point-to-point connection. Point-to-point connections are independent from one another and do not require a shared communications bus, which can become a bottleneck.
Now with X6, massive amounts of direct-attached Flash and spinning disk storage can be packed into an xSeries box, with memory-fast performance. Direct-attached storage (DAS), once considered an old-school means of storing data, has seen a revival in recent years for everything from high-performance databases used in data analytics to scale-out architectures used by cloud computing service providers.
Larger, high-memory servers can also be popular in virtualization shops because virtual machines that share a larger pool of resources encounter less contention, according to Bob Plankers, virtualization and cloud architect at a major Midwestern university.
Plankers' shop runs mostly Dell hardware, and while the X6 seems intriguing, he said, it probably won’t be enough to prompt a switch in vendors.
“You also have pitfalls to larger servers – it means you have a lot of eggs in one basket, too,” Plankers said.
Another IT professional said he liked what he heard of the new changes to the architecture, which could help the performance of one important application in his environment that is particularly disk I/O intensive. However, an aggressive move to flash storage may prove too expensive right now.
“At this point there is still too much of a financial premium to go to flash, so we wouldn’t be making the move with something like this now," said Nigel Fortlage CIO, Social Business Leader with GHY International, in Winnipeg Canada."But with engineering changes such as these, I would consider entertaining a technology like this in the next few years.”