LinuxGram

Voltaire, an InfiniBand loyalist through thick and thin – mostly thin for the last 10 years – has decided it’s long past time to stop beating its head against the wall and expand into Ethernet especially now that InfiniBand and Ethernet are starting to resemble each other and it can bring something to the party.

What it’s bringing is a scale-out Ethernet architecture – based on the next-generation Converged Enhanced Ethernet (CEE) technology that leverages characteristics inherent in InfiniBand – like a lossless fabric, cut-through switching and I/O virtualization – supporting what Voltaire likes to think of as “virtual mainframes” built out of industry-standard elements that make extensive use of virtualization (in other words, “the cloud”).

Actually Voltaire thinks Ethernet “stole” specs from InfiniBand but that’s another story. The so-called CEE standard hasn’t been etched in stone yet but Voltaire’s pretty sure where it’s going to wind up later this year. It can always reprogram if things change.

Anyway, it says its new widgetry, under development for more than 18 months, will improve the user experience and the economics of data center networking. It will also give Voltaire a less fussy crowd of potential users with less rigid performance requirements to cultivate than it’s had with InfiniBand.

Its new fabric is supposed to be a linearly scalable, low-latency, virtualized 10 gigE fabric with lower overall fabric costs and power consumption, greater efficiencies and simplified management even with thousands of nodes.

Voltaire’s approach flattens the expensive hierarchical switching tiers of a Cisco-based configuration that it calls “outdated,” “lacking in innovation” and a pathway to vendor lock-in.

The company claims that a customer building a 1,000-node data center with its magic can get 10x less latency and 4x faster core performance using 3x less switch power consumption for half the price of “alternatives.”

Voltaire says that means saving millions of the dollars.

And for good measure Voltaire says what it’s offering is an open standards architecture as opposed to what it calls Cisco’s Nexus lock-in, which “turns a data center into a Cisco appliance.”

Much of Voltaire’s rhetoric and positioning is now anti-Cisco and Cisco’s entrance into the server market has reportedly prodded Cisco’s competitors IBM and HP to cozy up to Voltaire, setting in train a fight between a standards-based solution and a proprietary one.

Voltaire got Merrill Lynch Research, now a Bank of America property, to opine that “Voltaire will likely be the first company in the market with a working Ethernet-based scalable solution. The company could utilize its vast experience in building grids or fabrics of servers, gained via years of building InfiniBand networks and we believe it may be the only company to have a truly scalable solution already in 2009 – far exceeding Cisco’s specs.”

Voltaire’s mojo is called a natural extension to its 20 Gb/s and 40 Gb/s InfiniBand switches and software and lets users choose their underlying data center fabric.

For its scale-out data center architecture Voltaire has built new high-density Layer 2 core switches that can be clustered to form large, linearly scaling Layer 2 fabrics. But one fabric can reportedly run LAN, IPS and storage traffic simultaneously with guaranteed service attributes and dynamic congestion management.

Its scheme puts L3-L7 switching and services at the edge on the theory that load balancing, say, isn’t needed on every switch.

So imagine L2 data center core switching as high-density, supporting many more nodes in a smaller footprint at a reduced budget; that allows large-scale L2 networks with a high bisectional Multipath bandwidth; that’s virtualization-aware, applying policies to VMs; and that’s multi-class, both lossless and lossy, enabling storage and IPC (lossless) with traditional LAN (lossy) while mitigating the congestion of lossless networks.

There’s also sine-qua-non fabric management software that controls all the physical and virtual switches and I/O devices in the fabric and does fabric monitoring and service-oriented policy enforcement.

Voltaire is not of course abandoning its InfiniBand heritage; it simply wants to see InfiniBand finally deliver on its original promise of being mainstream.

Last year the company’s revenues were up 16% to all of $62 million and it ended Q4 with $55 million in the bank but since going public in July of 2007 it’s only seen four profitable quarters. It wants that to change.

That’s why future software products have been designed to work on both InfiniBand and Ethernet.

The new chips are due later this year, when pricing will be available.

Cisco CEO John Chambers Monday became the latest in a long line of visionaries to come down from the mountain top with an architectural roadmap to the Promised Land at an exploitable industry inflection point.

If the purportedly game-changing next-generation platform scheme succeeds, Chambers will be rich, even more fiendishly rich than he already is; Cisco will own not only the data center, but the cloud, the data center’s logical successor, and the computer establishment, as we know it, will be up-ended.

As prophesied, the widgetry, Cisco’s answer to its slowing networking fortunes and the growth demanded of it by Wall Street, centers on a newfangled Nehalem-based blade server called the Unified Computing System, which is all about convergence with networking at its heart.

There’s no guarantee it will take off; it’s still slideware; and it will take time to gauge acceptance while Cisco wages what some call “rack-by-rack warfare.”

UCS unites compute, network, storage access and virtualization into a reportedly scalable and modular architecture that’s managed as a single system through Cisco’s new graphical USC Manager and its associated APIs for handling configurations and operations.

UCS has also adopted the idea of service profiles, sometimes called templates, to automate provisioning, but Cisco takes the template a step further and includes the firmware in the thing, which should hurry it up if nothing else.

Since Cisco has no history in servers, and since the UCS elements are often bought separately from different vendors, and since Cisco’s stepping all over the very toes that move tons of its routers and switches, the move is pretty gutsy, not to say cheeky considering Cisco’s merely a plumber.

HP, which owns the blade server market – well, 58% of it at any rate – and so stands to have the biggest black-and-blue toes, even more than IBM, claims there’s nothing new under the sun in Cisco’s architecture. It’s all already been done before and better, mostly by HP.

NCS was secretly developed under the codename Project California but HP’s director of strategy and architecture Gary Thome says it’s more like the Hotel California where, as the song says, “You can check out anytime you like, but you can never leave.”

In other words HP claims it’s pure vendor lock-in and that users will have to throw out the rest of their data centers if they go with the Cisco plan.

HP can’t imagine a data center scheme that doesn’t support Unix; figures its own BladeSystems can handle a wider variety of workloads than Cisco can; is pretty confident it can deliver headier power and cooling efficiencies; and says Cisco’s blade enclosure can’t function at all without its switch and that right there is a point of failure.

Paul Durzan, director of Cisco’s Server Access Business Unit, says HP’s talking mostly trash. Cisco boxes, for instance, can slip right into existing infrastructure without displacing what’s there.

Brocade, another competitor, said the same thing at HP: Cisco’s approach ain’t revolutionary; it’s capital intensive and – complex as the problem is that it’s trying to solve – doesn’t leverage open architectures and industry standards. Cisco of course resists the complexity charge since it’s reducing cabling, NICs and HBAs.

Anyway, since HP is out to cripple Cisco, it’s believed that HP, which used to move a lot of Cisco widgetry, will attack Cisco’s networking base even more than it has in the last few years – those Procurve switches, say – and that Cisco can kiss HP’s billion dollar contribution to its revenue stream good-bye.

Cisco appears not to care. There are far more billions to be made in data centers; Chambers, tempted to call it an “unlimited” opportunity, settled for saying it exposes Cisco to 25% of the $100 billion-a-year data center spend it’s never had a shot at before.

So as Cisco CTO Padmasree Warrior told the Wall Street Journal, “We are going to compete with HP. I don’t want to sugarcoat it. There is bound to be change in the landscape of who you compete with and who you partner with.” (One can almost hear agreements being shredded in the background.)

Cisco has dragged in a host of brand names to create a divide-and-conquer NCS ecosystem and accelerate its market adoption, folks like Intel (hey, it’s a new customer), VMware (Cisco owns a piece), EMC, BMC (providing the management software under a multi-year exclusive), Microsoft (well, nominally at any rate), Red Hat, Accenture, NetApp, Wipro, Oracle, Novell and Tata to name only a few.

Needless to say the names HP and IBM are not among them though Chambers told the Journal IBM is likely to be a partner too; at least it’s being cultivated but then IBM’s got Cisco rival Juniper in its pocket.

Apparently Cisco has offered to let IBM stick one of its blades in the UCS chassis and it doesn’t necessarily have to be an Intel blade. It could be a Power blade.

UCS is targeted at the big enterprise and service providers, and Cisco is harnessing 250 of its resellers, the ones who know something about the data center, to get there. Historically the channel has provided 80% of Cisco’s revenue.

The folks that’ll be buying this stuff aren’t Cisco’s usual end-user account managers. It’ll have to start in the C-suite with a concept sell and claims that CIOs may have to rejig their own internal organizations since NCS doesn’t square with the currently separate server buyer, storage buyer, networking buyer. Cisco muttered something about “unlocking the money in the cracks between the silos.”

The Cisco blades, otherwise known as the UCS B-Series, are supposed to be the start of a new family of Cisco products.

To make them special, they’re fitted with a patented extended memory that’s supposed to support applications with large data sets and allow significantly more virtual machines per server than usual.

See, UCS is supposed to be God’s gift to virtualization, “unleashing,” as Chambers likes to put it, virtualization’s “full potential” by enhancing the scalability, performance and operational control of virtual environments.

Cisco is supposed to have overcome the issues of security, policy enforcement and diagnostics that can hinder virtualization.

Each UCS system is supposed to be able to support thousands of virtual machines, promising to set a new high watermark for density.

Cisco and VMware, by the way, now have an OEM arrangement, and Cisco intends, among other things, to integrate VMware vCenter management suite. They mean to play in the cloud together, pushing virtualization down to the device and into the home.

There’s support for a “wire once” unified fabric over a low-latency, lossless 10 Gbit/s Ethernet foundation that consolidates LANs, SANs and HPC networks. This is supposed to reduce the number of network adapters, switches and cables needed and so lower both cost and energy.

As a matter of fact, although it has yet to say what it’s going to charge, Cisco claims that UCS will cut capital expenditures by 20% ands operational expenditures by 30%. Promises of cost savings are particularly timely right now. So is Cisco’s focus on the cloud, which is heavily dependent on networking.

Cisco says the unified fabric provides consolidated access to SANs and NAS over Ethernet, Fibre Channel, Fibre Channel over Ethernet and iSCSI, which ought to satisfy just about everybody.

Cisco owes NCS development to a start-up acquisition it made called Nuova Systems, which has reportedly been working on the blades for the last two years. What Cisco can do with the $30 billion it has it the bank – more than any other tech company – can only be guessed at, especially in a down economy.

“The key takeaway,” Chambers said at the end of his star-studded, mutual-grooming webcast Monday, “is it gives us a chance to perhaps become the leading company not just in communications but also in IT.”

NCS isn’t expected out until next quarter. It’s currently in beta at a reportedly10 beta sites such as Savvis and is deployed internally at Cisco.

There are eight blades to a UCS chassis and 40 chassis can be clustered into a 320-blade cluster. Cisco’s using the DP Nehalem.