LinuxGram

Amazon Web Services has started selling single-tenant instances that run on dedicated hardware inside an Amazon Virtual Private Cloud (VPC).

The classic definition of a cloud is multiple virtual machines on a server used by multiple tenants sharing the cost, but that definition doesn’t suit everybody’s security or governance needs so Amazon changed the definition.

It says the VPC can be just physically isolated Dedicated Instances or a mix of Dedicated Instances and non-dedicated instances.

Amazon is charging a premium hourly per-instance fee plus a $10-an-hour dedicated per-region fee regardless of how many Dedicated Instances a user is running there. The regional fee is supposed to compensate Amazon for business lost because the hardware isn’t multi-tenant.

There are hourly discounts on one- and three-year contracts.

If a user sets up multiple Dedicated Instances they may not all be on the same server; Amazon may spread them out, it says, to guard against hardware failure.

According to the rules, the tenancy of a VPC can’t be changed after it’s been created.

The pricing is spelled out at http://aws.amazon.com/dedicated-instances/.

Rackspace the other day started offering to support OpenStack, the eight-month-old open source cloud infrastructure platform it put together with NASA much to the chagrin of NASA buddy Eucalyptus Systems, the other open source cloud platform.

Support will come from a new business unit called Rackspace Cloud Builders that will also provide training, certification, management and deployment services to enterprises and service providers. The OpenStack widgetry is supposed to be ready to deploy at SP scale next month.

Rackspace said in a statement, “We are ready to support OpenStack deployments anywhere starting today.” What exactly it will cost is unclear. Presumably it’ll eventually add monitoring and management software from Cloudkick its other recent acquisition.

The support team consists of Rackspace developers and the 10 or so people from Anso Labs, the California-based professional services company that Rackspace just bought. Anso built the compute fabric under the Nebula mojo that NASA kicked into OpenStack. Rackspace supplied the more mature storage piece.

Anso co-founder Jesse Andrews is now director of development for Rackspace Cloud Builders and for all the cumbayá community malarkey Rackspace pretty much controls OpenStack. Even with the addition of three seats to the Project Oversight Committee (now the Project Policy Board) last week, Rackspace still controls seven of the 12.

Rackspace is apparently moving early in the game to justify the cost of OpenStack development despite the 50-odd partners the project has now gathered and to spur adoption. The widgetry recently reached the so-called Bexar cut, good for limited production runs, early last month and the more scaleable Cactus release is due in April.

Canonical, another Eucalyptus buddy, and Citrix are supposed to turn out software distributions.

Dell is supposed to bundle OpenStack with its servers, networking and services. It will be pushing customers to do a proof of concept and has built an OpenStack installer to turn bare metal servers into clouds that will be contributed to the community once field tests are run.

One of the first Cloud Builders clients is a collaborative project led by Cybera and Compute/Calcul Canada that is using OpenStack for a new program to reach out to Canadian small and medium-sized tech companies.

CANARIE, otherwise known as Canada’s Advanced Research and Innovation Network, recently launched the DAIR (Digital Accelerator for Innovation and Research) Program that leverages CANARIE’s high-speed network and adds compute and storage capabilities to deliver a shared R&D environment. To support the DAIR Program, Cybera and Compute/Calcul Canada are supplying and managing two large-scale compute and storage nodes using OpenStack as the pilot’s cloud service platform.

Rackspace says it is also working with Canonical, Citrix, Cloudscaling, Equinix, Intel, Microsoft and Opscode to deliver training, system integration and deployment services. Cisco has also recently signed up to contribute to OpenStack and integrate it with its own stuff.

It’s not every little start-up that can say that Intel built it its own personal processor but that’s what SeaMicro says happened when it needed a 64-bit Atom chip for its newfangled Atom-based servers.

SeaMicro CEO Andrew Feldman claims the 75-man company saw pretty enthusiastic adoption – it was “flooded with orders,” as he described it, for its first-generation 32-bit machine – enough for supply to have run short because the outfit hadn’t ordered enough of its proprietary ASICs, which had an 18-week lead time.

But there were potential customers that told the start-up to come back when it had a 64-bit machine. They couldn’t fit their data in 2GB of OS addressable memory and didn’t want to screw around recompiling their software to 32-bits.

So SeaMicro importuned Intel for the dingus and Intel built the low-power N570 dual-core processor for SeaMicro’s dense new SM10000-64 system, advertised as the world’s most energy efficient 64-bit x86 server, meant to underpin the modern data center and the ever-expanding cloud. It is the first Atom chip to support virtualization.

Intel didn’t even demand a huge order, suggesting that the semiconductor giant is more scared of ARM and its server aspirations than it lets on.

Armed with the dual-core 1.66GHz Atom, SeaMicro stuffed 256 of the doohickeys – 512 64-bit cores and 850GHz – into a 17.5-inch 10 rack unit-high system.

Besides putting a rack full of functionality in a 10U, it can put 2,048 cores into a seven-foot rack along with 1.024 terabytes of DDR3 DRAM, up to 64 SATA solid-state or hard disk drives and eight-64 one gigabit Ethernet uplinks.

It reduces power and space by 75%, cutting opex costs that really pinch. Google’s famous for complaining that servers cost more to power than to buy. The SeaMicro box is supposed to trash Amazon’s cloud pricing.

SeaMicro brags that the box delivers more compute per unit power and more compute per unit space than any x86 system ever built. It’s also supposed to be the industry leader in bandwidth per unit compute.

It eliminates layers of Ethernet switches, server management devices and expensive load balancers.

SeaMicro says its 1.28 terabit/second supercomputer-style fabric lets the SM10000-64 deliver five times more bandwidth per unit compute than traditional servers. That means that applications such as Hadoop that need extensive server-to-server communication can run more quickly by keeping network bottlenecks from inhibiting application performance.

Aside from the obvious off-the-shelf 64-bit operating systems support and x86 software compatibility – no modifications needed – the Intel chip can handle four gigabytes of operating system addressable memory per socket and four threads. On a SeaMicro motherboard – which is smaller than piece of copy paper – the N570 costs less than a watt for each gigahertz of compute at peak utilization with the company’s brand of power management.

A base configuration runs $148,000.

SeaMicro says the thing is already in production at sites it’s not allowed to divulge. It means to keep offering the 32-bit model, which is used by Mozilla, Skype, French Telecom, Rogers Communications and China Netcom Broadband. Japan is still locked into 32-bit. Anyway the CPUs can be swapped out.

Neither kit is any good at database, or CAD-CAM or supercomputer-y stuff but they are good at the web tier and with the LAMP stack.

They are supposed to make a big 3x leap in utilization, translating into $16 million in savings over three years, the company says, compared to a comparable Xeon setup.

Feldman expects SeaMicro to be profitable in the second half of next year.

There’s a provocative empty space in the middle of the company’s new motherboard for coming attractions in density.

By the way, SeaMicro’s servers are “Made in America” by Silicon Valley-based NBS.

The Committee on Foreign Investment in the US (CFIUS), the federal watchdog that reviews the national security implications of foreign investments in US companies and operations, told Huawei Technologies Monday night it should voluntarily unwind its sly 2010 acquisition of 3Leaf Systems, and Huawei said no.

Huawei is betting that the White House, where the decision goes next and can order divestiture, won’t do anything about it since it’s lobbying for US tech companies to get greater access to the Chinese market.

Its retort is altogether unusual. In fact, the Financial Times called it “virtually unprecedented.” The White House has roughly two weeks to act.

Huawei reportedly paid $2 million for 3Leaf’s IP and picked up 16 of its 50 people last May, failing to tell CFIUS anything about it and leaving CFIUS to discover it for itself months later.

CIFUS then mounted a Pentagon-demanded review, with Huawei arguing that, well, shucks, it didn’t buy the whole company. It left the buildings and equipment behind for 3Leaf’s creditors – even if it did buy 3Leaf’s servers out of bankruptcy – and so CFIUS didn’t need to look at it. Anyway, it told the Commerce Department about it. Of course it’s not the Commerce Department’s tuff; it can’t challenge such acquisitions.

The giant $28 billion Chinese maker of telecom and networking gear, started by a former officer in the People’s Liberation Army, has dubious connections and a bad rep with US authorities as a security risk.

Tainted by unsavory links to the Iranian Revolutionary Guard, the Taliban and the Chinese military-intelligence complex, its attempts to buy 3Com in 2008 and more recently to supply Sprint Nextel with gear and buy into Harbinger Capital’s high-speed wireless network have been blocked for fear the Chinese will be able to intercept US communications. Cisco sued it for stealing its router code and Motorola sued it for conspiring with its employees to steal its IP.

Huawei claims divestiture might further damage its brand and reputation and denies being tight with the Chinese establishment but on Thursday the Chinese government got into the act.

According to Bloomberg, Beijing’s Commerce Ministry told a press briefing that “As far as the investment activities of Chinese enterprises in the United States, it’s clear that there are many cases where the United States is using a security review to refuse investment by Chinese companies. You can say in some level it has influenced Sino-US cooperation. We also hope that the US can increase the transparency of the approval process and give Chinese companies investing in the US fair treatment.”

3Leaf, which, when last seen, depended on virtualization and its own special distributed cache coherency ASICs for Opteron chips, used to be able to make cheap racks of x86 blades and boxes into one big powerful scale-up shared-memory SMP system the likes of mainframes, proprietary mid-range machines and pricey RISC-based Unix boxes by merging their CPU power, memory and disk resources into pools it could construct and deconstruct at will.

Senators Jon Kyl (R-Arizona), Richard Burr (R-North Carolina) and Jim Webb (D-Virginia) and Representatives Illeana Ros-Lehtinen (R-Florida) and Sue Myrick (R-North Carolina) raised the issue of national security and the sanctity of US cloud networks with Huawei’s access to 3Leaf’s brand of cloud computing last week in a letter to Treasury Secretary Timothy Geithner and Commerce Secretary Gary Locke.

The start-up was backed by upwards of $67 million from Intel, among others, but hit the wall running last year when it couldn’t get additional financing.

If the White House goes ahead and orders divestiture, Huawei would have to sell 3Leaf’s patents.

Amazon may be so far upriver it can’t see this coming but there’s a start-up breaking cover that’s aiming to put a few shots across its bow.

It fancies it’s gonna leapfrog Rackspace and GoGrid to claim the number two spot in Infrastructure-as-a-Service (IaaS) behind Amazon in no time at all.

The name of this prodigy is NephoScale, a half-way erudite moniker from the Greek word nephos meaning, of course, cloud.

It was started by CEO Bruce Templeton, who bought Silicon Valley Web Hosting, NectarTech Hosting and Simpli Hosting with IPO money made when Foundry Networks went public, and CTO Telemachus Luu, who built GoGrid’s first three iterations, effectively making NephoScale his fourth cloud.

The start-up is internally financed. No venture capital until it proves its point, it says, and can get a happy valuation.

NephoScale claims it’s building a few tricks into its greenfield technology that not even the mighty Amazon has yet – like geographic redundancy of all systems, including provisioning, asset management, ticketing and billing – and that its stuff gives developers enough “unprecedented” control over the infrastructure that they will be able to go from development and testing on the platform to production on a global scale without breaking a sweat.

The platform’s reportedly unique architecture is supposed to be capable of scaling horizontally, vertically and geographically, simplifying the creation of complex computing environments. The architecture is supposed to avoid what the company calls the “design locks” or built-in limitations of rival cloud platforms like EC2′s lack of persistent storage and persistent IP addresses.

The start-up’s flagship services include virtualized cloud servers, on-demand dedicated servers and cloud storage à la Amazon’s S3, and Templeton says his team has built a programmatic interface called CloudScript that lets users provision and manage all aspects of the infrastructure using a single API submission, which gives the NephoScale platform PaaS capabilities.

He says, “We’re taking a new approach to IaaS by making it so simple for developers to autonomously set up a cloud infrastructure that it removes the system administrator from the equation, which means companies can do even more with less. The technology we’ve built into our architecture also helps us fulfill a real industry demand by supporting PaaS functions, such as configuring and optimizing an application stack, while providing the additional services and flexibility associated with an IaaS.”

CloudScript is supposed to put a level of infrastructure control into developers’ hands that they haven’t seen before.

Templeton says, “Imagine you’re a developer deploying 10 web servers with storage, firewalls and load balancers. Instead of doing individual API requests for each one, you can now do one API submission and call it done. You can begin to see how this can seriously impact efficiency and delivery times.”

The NephoScale widgetry has been in beta for eight months with a hundred testers, some household names, some Silicon Valley start-ups, some students. Templeton thinks 80% will turn into paying customers.

NephoScale’s cloud servers and cloud storage are available on a pay-as-you-go hourly basis, with no long-term commitment. Pricing, as detailed on its web site, is on a par with Amazon and the start-up is mimicking Amazon’s Reserved Instances with a membership plan that offers a 50% discount on the hourly rates of its Cloud Servers in exchange for an upfront annual fee ranging from $73 to $1,168. Its On-Demand Dedicated Servers will be billed month-to-month and can be had on an annualized pre-paid basis that includes two months free.

Like Amazon, the company also means to offer a free micro instance that unlike Amazon’s, which is only available to the user for a year, will be available indefinitely. Templeton claims to have a “secret sauce” that will make it cost-effective for NephoScale to do that.

NephoScale servers, all Intel stuff, can be had with 32- and 64-bit Linux and Windows operating systems. Currently that means CentOS, Debian, Ubuntu and Windows Server 2008, but the company expects to add Red Hat, SUSE, Fedora, Windows Server 2003 and SQL Server. Its Cloud Storage, which can scale to the hundreds of petabytes, is supposed to replicate objects three times, translating to uptimes of over Five9s and its Dedicated Servers, unlike, say, Rackspace, can reportedly be spun up in five minutes and scaled up and down by the hour.

The user will be able to see all of the widgetry in a single console.

Initially the start-up has a single data center on the West Coast but expects to add one in Asia this summer probably in Singapore and another in Europe by the end of this year probably in Frankfurt. It also expects to add facilities in the states.

The company isn’t quite all there yet with the functionality it means to roll out. For instance, Templeton says it will take another quarter or two for NephoScale to have the promised geographic redundancy in the bag as well as a promised geographically scalable single broadcast domain, effectively creating a private network.