Dell puts RNA Networks assets to use in new product

Dell has released a new software product designed to make solid-state-drive caching on servers more efficient, resulting in improved performance of applications such as databases.

The company's Fluid Cache 1.0 technology is the first product to be released with the technology acquired from memory virtualization company RNA Networks in June 2011. Dell was expected to implement RNA technology when it introduced its 12G servers last year, but no announcement came.

Dell has since moved slowly with RNA, taking more than a year-and-a-half to incorporate its technologies into servers. More products based on RNA will come in the future, a Dell spokeswoman said.

The Fluid Cache 1.0 software technology sits between the Express Flash solid-state drives attached directly to the PowerEdge 12G servers. The software essentially is an application accelerator that creates a high-speed caching pool to facilitate quicker reading and writing of data from SSDs.

The software layer replicates data in cache, which improves response time without compromising the data. Dell in a blog entry claimed that the software can boost database response times by up to 95 percent, and allows for the addition of more concurrent users accessing the database.

Caching technology has been prevalent in the form of read-only cache, said Kishore Gagrani, senior product manager at Dell, in a video explaining the technology that was posted on YouTube .

"Fluid Cache... can also rewrite cache. With the write caching comes data protection. We make sure data in the cache is protected," Gagrani said.

The software can scale caching performance as more Express Flash SSDs are attached to multiple servers. Fluid Cache 1.0 works only with Linux servers for now, but the company plans to add support for more operating systems such as Windows, Gagrani said. The technology will also be expanded to network-attached storage products, and will also be optimized for Compellent, which virtualizes storage.

Customers do not need to learn any new management tools, and applications do not need to be rewritten to work with Fluid Cache, Gagrani said.

Pricing and availability information for Fluid Cache wasn't immediately provided by Dell.

Agam Shah covers PCs, tablets, servers, chips and semiconductors for IDG News Service. Follow Agam on Twitter at @agamsh. Agam's e-mail address is agam_shah@idg.com

Dell testing 64-bit ARM server with chip from AppliedMicro

Dell has built a prototype server based on a 64-bit ARM processor from Applied Micro Circuits, which showed the system at a conference in Silicon Valley on Thursday.

Dell has already said it was testing servers based on 32-bit ARM chips from Marvel and Calxeda, but this is the first time it has shown any hardware based on a 64-bit ARM processor. Sixty-four-bit chips are generally better suited to server use than 32-bit parts.

Proponents say ARM chips will be more energy efficient than x86 processors that Intel makes for certain cloud and analytics workloads, but the market is in its early stages, with plenty of hardware and software development work to be done. Analysts estimate the first 64-bit ARM servers won't actually hit the market before 2014.

AppliedMicro hosted a session on Thursday at ARM's TechCon conference, where it tried to illustrate how various elements of the 64-bit ARM server "ecosystem" are coming together.

It was joined by representatives from Red Hat and Cloudera, both of whom said they'll have software ready for testing on 64-bit ARM chips next year. Oracle was also there, pledging a version of Java SE for 64-bit ARM processors, though it didn't give a timeframe.

AppliedMicro CEO Paramesh Gopi, in full showman mode, pulled away a black cloth cover to reveal the Dell server at the end of his talk. He didn't describe it in any detail but it appeared to be a two-rack-unit chassis with four or five individual servers, or "sleds," that slide into the frame.

The hardware was a prototype, and it's still unknown if Dell will actually sell an ARM-based server using AppliedMicro technology. Dell is experimenting with ARM components from several suppliers, and it was also at AMD's event Monday when it announced plans to build ARM-based server chips.

"We don't have any plans to make generally available an ARM-based server right now -- that includes the Applied Micro-based prototype you saw," Dell spokeswoman Erin Zehr said via email. "We're currently focused on ecosystem enablement -- giving developers access to clusters so they can test or write to ARM," she said.

The processor inside the Dell system, which AppliedMicro called an "X-Gene" processor, was also an early prototype. Gopi said X-Gene parts will be ready for customers to begin testing in the first quarter next year, with commercial products coming later in 2013.

But AppliedMicro does now have actual prototype silicon, which is a step up from the HotChips conference in August, when it showed a server board with a mock-up chip.

It demonstrated its hardware in action Thursday. It showed a website running on what Gopi said was a prototype X-Gene server built by AppliedMicro and located in a remote data center. He streamed a trailer for the new James Bond film, which appeared to run smoothly.

"We are literally months away, ladies and gentlemen," he said. "In Q1 next year, you'll have not only silicon but also the software I just showed you and systems to go around it." He was still referring to prototype systems, however.

Gopi also unveiled three server reference designs that AppliedMicro has come up with, to show server makers what they can build. They're dubbed X-Memory, X-Compute and X-Storage, depending on the target application.

The X-Storage system is aimed at Hadoop-type analytics applications, and combines a sea of hard disks with a single X-Gene server board. It had a total 36TB of storage, Gopi said.

ARM offers two types of licenses for its chip designs. Companies can buy an architectural license, as AppliedMicro did, and design their own processor from scratch. That allows for greater customization, but takes more time and money. They can also buy a license for a finished processor design.

ARM unveiled its 64-bit architecture, ARMv8, at last year's TechCon. The news earlier this week was that ARM has now released its first 64-bit processor designs, the Cortex-A57 and A53. Chips based on those designs could appear by the end of this year.

James Niccolai covers data centers and general technology news for IDG News Service. Follow James on Twitter at @jniccolai. James's e-mail address is james_niccolai@idg.com

Dell supercomputer effort spawns new line of servers

Dell has developed a line of servers based on designs the company is using in an upcoming 10-petaflop supercomputer called Stampede, which will be fully deployed at the University of Texas, Austin, starting next year.

The PowerEdge C8000 servers are built with standard Intel x86 CPUs and can be equipped with graphics processors or additional storage to improve performance on database tasks, high-performance computing operations and cloud workloads.

Users will be able to mix and match graphics processors, storage, memory and other elements inside the servers, said Armando Acosta, a product manager at Dell.

For its part, the Stampede supercomputer includes thousands of C8000 servers with a total of 272TB of memory and 14 petabytes of storage. Dell and the Texas Advanced Computing Center at the University of Texas worked together on Stampede. The design for the C8000 servers blossomed as the supercomputer came to fruition, Acosta said.

The supercomputer will use eight-core Intel Xeon E5-2600 processors and co-processors code-named Knights Corner, which Dell said will speed up scientific and math calculations.

As for the new servers, the basic C8220 chassis can have up to eight blade servers; each server can contain two CPUs with up to 16 processing cores, two internal hard drives and additional storage and networking options. For instance, the servers can be hooked up to the new C8000XD storage box for expandable hard drive or SSD options.

The C8220X, a more advanced model in the new lineup, has more RAM and storage and can be equipped with graphics processors. All of the servers are designed for use in highly parallel computing environments, Acosta said.

Pricing starts at $35,000 for the C8220, $42,000 for the C8220X and about $25,000 for the C8000XD storage box.

This version of this story was originally published in Computerworld's print edition. It was adapted froman article that appeared earlier on Computerworld.com.

Read more about hardware in Computerworld's Hardware Topic Center.

IBM drops Power7+ in high-end Unix servers

IBM has started to roll out a new processor for its Power family of servers, a staggered affair that will start with higher-end systems and eventually reach the midrange and low-end boxes.

The new Power7+ chip has a higher clock speed than its predecessor, at up to 4.4Ghz, but the biggest change is in the Level 3 on-chip memory cache, which IBM has expanded to a sizeable 80MB, from 32MB on the Power7.

The bigger cache means more of the data being used for calculations -- the "working set" -- can be stored on the chip close to the CPU cores, which helps to speed operations. With a smaller cache, data has to be fetched more frequently from main memory.

The higher clock speed and larger cache will give a boost in performance for databases and Java applications, according to Satya Sharma , CTO for IBM's Power Systems business and an IBM "fellow," or one of its top engineers. "We can improve performance for some Java applications by up to 40 percent compared to the Power7," Sharma said.

IBM's top brass are due to discuss its systems business during a customer webcast at 11 am Eastern Wednesday. They may also talk about a new, high-end storage system called the DS8870 and an update to IBM's DB2 Analytics Accelerator, which are also being announced.

Across the country at about the same time, Oracle systems chief John Fowler is due to give a keynote speech at Oracle OpenWorld in San Francisco, and Hewlett-Packard is holding its financial analyst day, where it's sure to give an update on its systems strategy.

None of the vendors have a lot to cheer about, at least when it comes to Unix: Unix systems revenue dropped 20 percent in the June quarter, to $2.3 billion, according to recent figures from IDC. But at least IBM's revenue declined only 10 percent, and it managed to gain 6.1 points of market share, IDC said.

The Power7+ is being offered now for the Power 770 and 780 systems, which sit near the high-end of the Power line-up. It will come eventually to lower-end systems as well, like the 740 and 750, but IBM isn't saying yet when that will be.

IBM's most powerful Unix machine, the Power 795, won't get the new chip at all, Sharma said. Customers that buy such high-end systems generally prefer stability to incremental upgrades, he said, adding that IBM took the same tack when it introduced the Power 6+.

The 795 will get at least one technical boost, however. IBM is introducing a new memory module with twice the density, so the maximum configuration for the 795 increases from 8TB of main memory to 16TB. The 770 and 780 also get the denser DIMMs.

IBM is also introducing a new memory compression accelerator that can "make a 32GB system look like a 48GB or 64GB system," Sharma said. That can help reduce memory costs for customers, but there's a trade-off in increased latency as data is decompressed for use.

With the Power7+, IBM has also doubled the number of virtual machines customers can run on each processor core, to 10 VMs. While customers might not want that many virtual machines for production use, developers can use them for jobs like compiling code, Sharma said.

Another feature that's been discussed with the Power7+ -- the ability to put two processors in one socket -- also isn't available yet. The DCM, or Dual Chip Module, effectively increases the operations per second that customers get from each socket, with the trade-off that the cores run at lower clock speeds.

It's not being offered for the high-end Power systems machines, however. "It's partly that there is a little bit more work to do, and partly it's the class of systems where we want to use that capability," Sharma said. He wouldn't give details but implied the technology is destined for midrange or lower-end systems like the Power 740 and 750.

IBM offers a sort of "compute on demand" scheme for its Power systems, through which customers can pay to activate additional processor cores for a few weeks or months, such as during the holiday shopping season, then disable again them afterwards.

It's running a "special offer" for customers who buy a new 780 or 795 system. For each processor that ships with the server, they get 15 days of additional processing for free. So if a customer buys a Power 780 system with 16 cores enabled, they also get 15 days when they can turn on an additional 16 cores.

It's also introducing a concept called Power System Pools, which lets customers use those 15 days -- and any other compute days they purchase -- across up to 10 different 780 or 795 servers. Essentially it gives customers more flexibility in how they allocate their processor resources, and, IBM hopes, makes them a bit more likely to choose IBM.

IBM doesn't usually publish prices for such high-end systems but the faster chips will come in at about the same price point as their predecessors, Sharma said. It will offer slightly better pricing on the Power 780 because it would prefer more customers bought its higher-end systems, he said.

James Niccolai covers data centers and general technology news for IDG News Service. Follow James on Twitter at @jniccolai. James's e-mail address is james_niccolai@idg.com