Mellanox Builds Bridge to Data Center Consolidation

Mellanox Technologies Ltd. today rolled out another alternative for data center consolidation, a gateway that lets organizations run Fibre Channel (FC), Ethernet and InfiniBand on one network.

The BridgeX gateway connects Ethernet or InfiniBand switches to Ethernet or FC networks. Mellanox sells InfiniBand switches, silicon and host bus connectors (HBCs), and last year began diversifying its product platform to try and provide one fabric to connect the entire data center.

BridgeX supports the Virtual Protocol Interconnect (VPI) technology that Mellanox launched last year for its ConnectX host adapters that support InfiniBand, Ethernet and Fibre Channel over Ethernet (FCoE). VPI supports port auto-sensing of the fabric, and configures the adapter to the correct mode.

The BridgeX gateway comes as Brocade, Cisco Systems Inc. and other network connectivity vendors are putting together an ecosystem for FCoE to consolidate storage-area networks (SANs) and local-area networks (LANs). Cisco supports FCoE in its Nexus switches, while Brocade plans to enable FCoE through its DCX Backbone platform. Fibre Channel host bus adapter (HBA) vendors Emulex Corp. and QLogic Corp. have launched converged network adapters (CNAs) for FCoE, and storage array vendors EMC Corp. and NetApp have pledged FCoE support in future systems.

BridgeX can serve as an alternative to FCoE or enable it – Mellanox will sell BridgeX silicon to Ethernet switch vendors to develop FCoE switches. BridgeX devices do not require InfiniBand or ConnectX cards.

BridgeX supports 40 Gbps InfiniBand, 10 Gigabit Ethernet and 8 Gbps Fibre Channel. It can be used as an InfiniBand-to-Ethernet and FC gateway, or Ethernet-to-Ethernet and FC gateway. The initial product in the platform, the BX4000, has four InfiniBand or 12 Ethernet uplink ports, and up to 16 FC or 12 Ethernet downlink ports. When the uplink ports are Ethernet, BridgeX can serve as an FCoE link to Fibre Channel storage.

BridgeX promises many of the same advantages that storage networking vendors claim for FCoE – it can reduce cabling, power and cooling in the data center, and also gives administrators one fabric to manage. Gilad Shainer, Mellanox’s director of technical marketing, said BridgeX can enable consolidation without requiring a major data center upgrade.

“Other vendors are forcing people to use a specific solution,” Shainer said. “We’re saying, ‘How can we enable you to use all the applications and transports you’re using now?’”

While most industry experts say FCoE or other methods of data center consolidation probably won’t happen on a widespread basis until 2010 or later, analysts say enterprises are beginning to form their strategies now.

“I think converged networks are inevitable,” said Nik Simpson, senior analyst at Midvale, Utah-based Burton Group. “There are just too many advantages to it from an overall view of the enterprise to see it not happening. We’re going to see more and more storage arrays with native FCoE targets, and we’re seeing server platforms with 10 gigE on the motherboard.”

Simpson said BridgeX is a good way to consolidate cabling without doing a forklift upgrade.

Zeus Kerravala, senior vice president of global enterprise research at Boston-based Yankee Group, says it will probably take a while for a dominant consolidation technology or protocol to emerge, but BridgeX can serve as an early step in the process.

“This provides a roadmap to FC over Ethernet,” Kerravala said. “It offers companies a way to try it out or at least phase it in. This is a way for companies to start thinking of consolidation without having to commit a lot of capital to it because it doesn’t have to be a hard rip and replace.”

Other vendors have offered InfiniBand gateways for FC and Ethernet, but Mellanox’s Shainer said they had to first terminate InfiniBand and then initiate a new protocol. He said that added cost and power requirements, and prevented them from working at wire speed.

Pricing for the BridgeX gateways starts at $9,995, and the silicon costs less than $500 per box. Shainer said he expects server vendor OEM partners to begin shipping in volume next month.

Surgient’s Virtual Datacenter Updated with Woven Systems’ 10 Gig

Surgient, a company that provides self-service virtualization automation and lab management software, provides its technology to customers in two ways: a licensed model and a hosted model. In its hosted model, the company makes its platform available as part of a hosted service that enables businesses to save considerable resources and slash time-to-market for new software products with a hosted datacenter for testing, training, or demonstrating their software.

Surgient’s virtual datacenter leverages virtualization technology from Microsoft and VMware. And as the company’s list of hosted customers continues to grow, so too must its datacenter. Even though its datacenter is itself virtualized, the company has to deal with increased scale without compromising performance.

To help solve that problem, Surgient looked outside of its organization and found Woven Systems and their Ethernet fabric switching solutions based on the company’s patented vSCALE technology for datacenters and high-performance computing clusters.

Surgient chose Woven’s technology to upgrade its datacenter to 10 Gigabit Ethernet using Woven’s EFX 1000 Ethernet Fabric Switch. With only a single EFX 1000 switch, Surgient is able to interconnect all of its virtualized servers and its storage area network in its datacenter. So why did Surgient choose Woven Systems?

“We looked at Ethernet switches from multiple providers, and Woven’s Ethernet fabric provided the most impressive performance,” said Evan Watkins, director of operations at Surgient.

“Dynamic datacenter environments that make extensive use of virtualization have dynamic networking requirements. As virtual machines are moved around to different virtualization hosts as part of normal operational tasks, the networking load shifts with them, changing the load on various switches in the environment. In a traditional networking environment using spanning tree protocol, it is normal to manually plan that load shift, making sure that the intermediate distribution layer switches have sufficient bandwidth and uplink capacity to handle the loads. With the Woven Systems EFX-1000, it dynamically adjusts the load on the various paths in its internal fabric, avoiding bottlenecks and optimizing throughput without any operational efforts on our part.”

In a typical virtualized datacenter, it is critical to have a reliable SAN environment. However, it can prove to be extremely expensive for most organizations to build out a Fibre channel storage network system along with a Gigabit network environment. And Surgient’s hosted facility is no exception.

Watkins said, “With the Woven Systems switch we have a reliable, high speed, multi-path, load balancing iSCSI storage network but only have one type of network to manage. We didn’t have to deploy a Fibre channel storage network. We didn’t have to hire or build Fibre channel networking skills on our datacenter team. This saved us a considerable amount in capital and operational expenses.”

“Surgient’s deployment proves it is now possible to build an all-Ethernet datacenter that scales without compromising performance,” says Joseph Ammirato, vice president of marketing at Woven. “They’ve amassed an impressive and growing list of customers that demand the mission-critical reliability afforded by Woven’s Ethernet fabric.”