Voltaire has added Ethernet switch management to the Infiniband fabric management product, Unified Fabric Manager (UFM). UFM provides the management functions to provision the switching fabric from port to port and manage congestion within the fabric. In addition, UFM provides reporting and prediction tools to determine where bottleneck occur and manages capacity before it becomes a problem. Voltaire also integrated UFM with third party products like Blade Networks G8124 and Procurve's 6600-24XG top of rack switches and NICs from Chelsio, Intel and Mellanox.
In 2009, Voltaire entered the 10Gb data center switching market with their 8500 switch and hoped to leverage their presence in the Infiniband market to enter the 10Gb Ethernet market competing with the likes of Arista for high speed, low latency networking. It's a tough market to enter, Steve Schuchart, Principal Analyst with Current Analysis thinks. "Voltaire has not generated any real buzz and is being ignored by competitors, which isn't a good sign. Between being only Layer 2 and being an unknown startup, any relevance Voltaire has will be limited to the HPC market," he said.
Unlike the fight for the campus LAN where the battle rages over cost per port, the data center network battle is about high capacity, flexibility and reliability. Network problems in the data center affect applications that affects everyone. While Cisco dominates in the LAN, the data center is a more open field where some products are a better fit than others. In high-performance situations where low latency is a critical requirement, such as financial trading or data modeling where processing is highly transactional, the difference in microseconds can have an impact on whether a trade is completed or the time to crunch data. Infiniband vendors like Voltaire build switching platforms for these transactions.
Fabric is one of those vague terms that describes the network paths from server port to server port and includes services that enforce service level agreements, quality of service, congestion control, multiple paths and all the management and reporting functions that enable efficient and robust network from any point to any point. Fabric management will become increasingly important in data centers as companies move from a data center architecture where physical and virtual servers are in relatively fixed locations to a more dynamic architecture where physical and virtual servers can appear and disappear from the network based on demand, availability, and disaster recovery.
You can expect to hear more about each vendor's fabrics in the coming years as they ramp up their product lines to meet data center demands.