LinuxGram

SeaMicro’s back. Four months after it pushed out its revolutionary low-power 64-bit Atom server it’s filled the provocative empty space in the middle of the thing’s five-by-11-inch motherboard to make the dingus denser.

The brand new SM10000-64HD – the start-up suffers from the same unimaginative naming conventions as Intel – stuffs 384 dual-core 1.66GHz Atom chips in a box.

That’s 768 64-bit cores worth 1,275GHz in a 10U Made-in-America system, or 3,072 Intel cores in a seven-foot rack.

It calls it simply the “the world’s most energy-efficient 64-bit x86 server,” consuming all of 3.5kW under typical workloads.

It has improved the compute density of the original SM10000-64 by 150% and increased its industry-leading compute-per-watt metric by 20%.

Despite its density this generation takes less power to run than older brother through the practice of some arcane black magic.

SeaMicro, which claims breakthroughs in CPU design, virtualization, supercomputing and networking, says it now delivers more compute per-unit power and more compute per-unit space than any x86 system ever built, offering the industry a new solution for the power and space crises that eat up 75% of users’ opex.

The new box replaces 60 traditional servers, four top-of-rack switches, four terminal servers and a load balancer while using a quarter of the power and a sixth the space at a quarter the weight without any changes to x86 software.

The company calculates it can save $3.3 million-$27.4 million in power and real estate and reduce TCO by more than 80%.

The SM10000-64HD, like its forerunner the SM10000-64, is based on the N570 Atom chip that Intel built especially for SeaMicro.

Intel, in turn, is gleefully surprised to be a few steps ahead of ARM, which spooks the…well, worries it. And, while the Atom ain’t the pricey high-margin Xeon, according to CEO Andrew Feldman, SeaMicro’s many-Atom box gives Intel its highest profit solution in a rack.

Intel calls it simply “an amazing accomplishment.”

The N570 supports four gigs of operating system addressable memory per socket and is the first low-power Atom to support virtualization. Each dual-core N570 also supports four threads and is said to deliver the industry’s best performance per-watt for Internet workloads. When used on a SeaMicro motherboard in conjunction with SeaMicro power management technology the N570 uses, at peak utilization, less than one watt for each gigahertz of compute.

SeaMicro’s architecture simplifies data center operations and management by eliminating layers of Ethernet switches, server management devices and expensive load balancers. It’s plug and play – customers can deploy it without modifying existing operating systems, application software or management tools.

It accommodates 1,536 terabytes of DDR3 DRAM, up to 64 SATA solid state or hard disk drives and eight-64 one gigabit Ethernet uplinks.

SeaMicro’s scale-out systems are optimized for the 64-bit Web 2.0 and cloud data centers that are spreading everywhere. They are also being used by telcos, the US government and big enterprises.

The four-month-old SM10000-64 passed a million hours in production the week before last, Feldman said.

Available now, the SM10000-64HD starts at $237,000.

SeaMicro, which is backed to the tune of $60 million from Khosla Ventures, Draper Fisher Jurvetson, Crosslink Capital and four mystery strategic investors plus $9.3 million in Energy Department grant money, expects to be profitable by this time next year.

Sales have been doubling quarter-on-quarter with customers on five continents and SeaMicro thinks it could qualify as the fastest-growing systems company in the history of Silicon Valley.

Customers include France Telecom, Skype, Mozilla, eHarmony, China Netcom BB and a lot of other unmentionables. Although it complains that servers cost more to power than to buy, Google is still too not-invented-here to give it a toss.

NYSE Technologies, the two-year-old technology division of NYSE Euronext, said Wednesday morning that it’s going into business with EMC and VMware to offer cloud services such as historical market data, algorithm testing and analytics to Wall Street firms that have previously used their own servers for such things.

The venture is called the Capital Markets Community Platform, still another kind of cloud, a “special-purpose or community cloud,” dedicated solely to the financial services industry and meant to take massive costs out of the system.

It’ll also give NYSE Euronext something else to do with its shiny new data centers in New Jersey and Basildon outside of London which will provide dedicated boxes to those who insist. NYSE Technologies is currently generating revenues of $460 million and wants to get to a billion by 2015.

The exchange moved all trading systems for the NYSE and Amex stock markets to Mahwah last year and that space is being sold to hedge funds, banks and trading firms looking to get their servers as close to the action as possible to maximize the speed at which trades are executed.

The Capital Markets Community Platform is apparently supposed to aggregate data from exchanges, broker-dealers and the buy side.

VMware is kicking in vSphere, vCloud Director and vShield security as well as the EMC VNX series of unified storage platforms leveraging EMC FAST VP and FAST Cache.

The widgetry, which already has two customers who don’t mind their names being used – Goldman Sachs unit Pico Quantitative Trading and hedge fund Millennium Management – is ostensibly supposed to go live July 1. Till then it’s in beta.

There will be other hubs in Sao Paolo, Tokyo and Toronto this year.

Facebook has been quietly working on building an energy-efficient data center at the lowest possible cost for a year or two and Thursday it open sourced the infrastructure it created under what it calls the Open Compute Project, figuring that many eyes might improve on it.

“We are not the only ones who need the kind of hardware that we are building out,” Facebook CEO Mark Zuckerberg said.

The company wasn’t getting what it needed from the OEMs so it designed the thing from the ground up and wound up building its own stripped-down Intel and AMD servers, power supplies, server racks and battery backup systems.

Its approach meant it had total control over every part of the system from the software on up and that meant it could use a 480-volt electrical distribution system to reduce energy loss; remove anything in the servers that didn’t contribute to efficiency; reuse hot aisle air in winter to heat its offices; and eliminate the need for a central uninterruptible power supply.

As a result it claims its soon-to-open data center in Prineville, Oregon is 38% more efficient and 24% less expensive to build and run than other state-of-the-art data centers. It’s also supposed to support more Facebook users and offer them real-time social experiences – such as the ability to read comments as they are written or see friends of friends appear dynamically as the user searches.

The facility is reportedly fetching a PUE of 1.07, which is below the EPA-defined state-of-the-art industry average of 1.5 and means 93% of the energy from the grid makes it into every so-called Open Compute server.

It’s eliminated centralized chillers and 480V-to-208V transformation. It says Ethernet-powered LED lighting and its passive cooling infrastructure cut the energy spent on running the facility further.

The servers it designed reportedly use 22% fewer materials and are 1.5U high to accommodate taller heat sinks and larger fans. The power supplies, evidently from HP, are reportedly 94.5% efficient.

Thursday it released the specifications and mechanical drawings covering its motherboards, power supply, server chassis, server racks and battery cabinets. It’s also sharing its data center electrical and mechanical construction specifications.

Greenpeace, a critic of Facebook’s piggish energy use, says efficiency isn’t enough. It needs to tap clean, renewable energy sources instead of coal.

Facebook worked with Alfa Tech, AMD, Delta, Intel, Power-One and Quanta to develop the first generation of the technologies used and is working with Dell, HP, Rackspace, Skype and Zynga on its next generation. Apparently it’s not thinking of using ARM.

It says it’s already talking its peers about how they can work together on Open Compute Project technology. Competitors were seen at its announcement.

See the server specs and mechanical designs and the data center specs and mechanical designs.

Deutsche Telekom, Facebook, Google, Microsoft, Verizon and Yahoo, all big network owners, said Monday that they have formed a standards-setting Open Networking Foundation (ONF) that’ll promote an approach to networking called Software-Defined Networking (SDN) that makes networks programmable like computers.

They got 17 other companies to sign up including heavyweights like Cisco, Broadcom, Juniper Networks, NTT, IBM, HP and VMware.

ONF says SDN works through relatively simple software changes and applies to all kinds of networks including data centers, wide area telecommunication networks, wireless networks, enterprises and homes.

SDN is supposed to give network owners and operators better control over their networks, let them optimize network behavior and prioritize data.

ONF said SDN can be used in data centers to reduce energy usage by allowing some routers to be powered down during off-peak periods. It could also set up on-demand “express lanes” for time-sensitive voice and data traffic or let telecom giants combine several fiber optic backbones together temporarily to handle heavy traffic, the New York Times said.

It’s a matter of newfangled centralized cloud computing versus old-line decentralized network design, the paper said. It separates packet switching mechanisms from control functions.

SDN comes out of six years of research at Stanford and Berkeley. It’s based on a software interface called OpenFlow for controlling how packets are forwarded through network switches (sorta like the BIOS firmware in a PC, GigaOM says), and a set of global management interfaces that advanced management tools can be built on. It’s supposed to eventually improve security and might improve privacy.

It could perhaps – blessedly – detect DDOS attacks better.

GigaOM describes it as the commoditization of networking imagining Google buying networking silicon from Broadcom and building its own switches, creating its own network topography using OpenFlow and putting firmware providers like Cisco, Juniper and Force 10 at risk.

ONF says its first job will be to adopt and lead the development of the OpenFlow standard (www.openflow.org) and encourage its adoption by freely licensing it to all member companies, hoping for supporting hardware and controllers by the end of the year.

It will then start defining global management interfaces.

Google’s senior VP of engineering Urs Hoelzle will be president and chairman of ONF. The general manager of Windows Azure Infrastructure at Microsoft Arne Josefsberg is on the board.

The rest of the members include Brocade, Ciena, Citrix, Dell, Ericsson, Force10, Marvell, NEC, Netgear and Riverbed Technology.

Cisco, HP and Juniper have prototypes supporting OpenFlow.

See www.opennetworkingfoundation.org.

Fasten your seatbelts, folks, it may get to be a bumpy ride.

Very late Friday – 9:30 at night in New York in fact, when Japan was dominating the news – Lawson Software said it had gotten an unsolicited $1.84 billion takeover offer from privately held Infor and its owner Golden Gate Capital.

That was three days after Reuters reported on Tuesday that Lawson had hired Barclays Capital to assess its options.

Reuters ticked off its options as being HP, IBM, Microsoft, the avaricious Oracle, SAP, privately owned Infor and private equity firms. You can cross a couple of those names like IBM and Microsoft off on principle. IBM buying application software would be a change in direction.

Anyway, the important thing to remember about Infor is that ex-Oracle co-president Chuck Phillips is CEO there now, having been tossed out of Oracle to make room for ex-HP CEO Mark Hurd after Phillips had been embarrassed by some indiscreet billboards advertising his extramarital love life a few months before.

Larry Ellison watchers assume that because Infor is suddenly interested in Lawson that Oracle’s CEO may be tempted to up the ante. And Phillips, they say, probably wants Lawson even more than he otherwise might because he knows Larry might want it.

Larry of course has a ton of cash, Lawson will be accretive anyway, buying it will add to Oracle’s competitive mass against SAP, and Larry probably wouldn’t mind paying a “spite premium” to leave Chuck standing at the altar.

That’s if Larry thinks that the price Infor is offering is fair enough to justify the spite premium. If not, he’s likely to say publicly that Chuck is foolishly overpaying for the place.

Ellison also probably wouldn’t mind playing a round of “keep away” with ex-Oracle president-now-non-executive chairman of HP Ray Lane as a bonus, so it is already makes some sense to buy it.

HP of course is under considerable pressure to add to its software position. HP’s new CEO Leo Apotheker, the ex-CEO of SAP and a guy whose reputation Ellison has already tried to muddy recently, is expected to say so on Monday.

Infor’s cash bid works out to $11.25 a share. Lawson’s stock closed at $11.55 Friday, a relatively airy height for the multi-category, multi-industry enterprise software house and 2.5% more than Infor offered.

Last Monday, before Reuters reported it was in play, Lawson stock was at $9.88. Bloomberg figures its average price over the last 90 days has been $9.46, making the Infor offer a 19% premium. Bloomberg also notes that the average premium the last three years for software acquisitions over a billion dollars has been 30.3%.

Lawson said in its disclosure of the Infor bid that it was in discussions with Infor but warned that “there can be no assurance that any agreement will be reached,” also indicating that Barclays – which it now admits hiring – may be trolling for alternatives.

Lawson held out the amusing possibility it might not sell at all. Who are we kidding here? Activist shareholder Carl Icahn owns around 10% of Lawson and will push to sell to the highest bidder.

Lawson, which did $736 million in revenue its last fiscal year ended in May, said it “does not intend to comment further regarding this matter unless and until an agreement is reached, discussions with Infor and Golden Gate have been terminated or the board concludes its strategic review.” Reuters believed Tuesday that the so-called strategic review was in its early stages.