Long involved in the discussion about and the market for converged I/O, Xsigo wants to be part of a larger debate and a potentially much bigger market opportunity.
Xsigo said last summer that its goal was to virtualize components of data-center networking, just as servers and storage have been virtualized previously. Wait, some of you might say, isn’t that the purview of software-defined networking (SDN) vendors? Well, yes, that’s true, and while there are obvious differences between what Xsigo delivers and what’s being put on the table by SDN purveyors, Xsigo thinks it has a compelling story to tell.
Xsigo’s I/O Director started off addressing virtualization and data transfer between servers and storage. Last summer, though, its I/O Director stepped up to the server-to-server challenge, simultaneously extending its incursion onto server turf while making a claim on networking territory.
Cisco Takes Notice
That got the attention of Cisco Systems, which offers networking and servers, and a relatively vehement vendetta ensued between the two companies. Xsigo probably got more benefit than Cisco did from the mutual antagonism, if only because Cisco’s public reaction to Xsigo indicated that the smaller player had done enough damage to be considered a threat by the networking giant. In aiming its competitive marketing guns at Xsigo and blasting away, Cisco explicitly acknowledged Xsigo and implicitly conferred added legitimacy in the process.
At any rate, with the addition of the Xsigo Server Fabric, which began shipping in earnest toward the end of last year, the Xsigo I/O Director now allows servers and devices to connect to each other directly without going over the network. As a result, adding a virtual machine (VM) doesn’t involve using an IP address or setting up a virtual LAN (VLAN). That’s addressed by I/O director and its virtual server interfaces.
Market analyst Zeus Kerravla has said that the Xsigo Server Fabric creates a new infrastructure atop the physical network, which is true enough. The Xsigo Server Fabric obviates the access-layer network, allowing servers and their VMs to communicate directly.
Bumping Layers
Xsigo contends its Server Fabric also effectively eliminates the aggregation layer. Xsigo says its infrastructure extends as for as the core network, where it is compatible with switches from any of the major players, including Cisco and Juniper. As such, Xsigo says its technology transforms a hierarchical network into a pool of bandwidth that can be used to connect virtualized resources in a data center.
By reducing the numbers of switch ports and infrastructure layers — the company says there’s just one layer of connectivity management between the OS or hypervisor and the core network with its approach as compared to as many as four layers in the Cisco model — Xsigo says its business model is the exact opposite of Cisco’s. Further to that point, Xsigo says that it is open, acting as a transparent conduit moving data between servers and the network core, whereas it alleges Cisco is not. Finally, Xsigo says it has no server agenda, whereas Cisco pushes its own servers as part of its Unified Computing System (UCS) for data-center virtualization.
Playing Its Part
Having no server agenda and taking a cut of the networking pie seem to have resulted in a go-it-alone strategy for Xsigo. It’s conceivable that market dynamics and shifting vendor alliances could change that picture, but for now Xsigo doesn’t have a powerful technology-partner ecosystem to leverage. As The Register noted, Xsigo has no OEM deals and is not thought to be an acquisition target of a major player, though Dell is responsible for about 20 percent of Xsigo’s sales and Oracle is cited as a potential acquirer in some quarters.
Xsigo customers, including some big names, have derived some significant cost savings from cutting down on cabling and getting much greater utilization from servers, virtual machines, and their network resources.
While not a member of the SDN fraternity, Xsigo wants us to know that it is playing its part in virtualized infrastructure for the data center.
Twilight in the Valley of the Nerds 
“Virtualized infrastructure” covers a lot of ground, and we’ll see many competing solutions in the market.
At Xsigo, we believe there’s definitely a need for what we do: an open solution — one that interoperates with anyone’s servers, networks, and storage. And a solution that’s proven to converge Fibre Channel and Ethernet to a single infrastructure. (Xsigo has 250 customers in production, most of whom use Fibre Channel.) And one with 40Gb performance now, at the same cost as 10Gb. (and at about the same power!)
The benefits of a virtualized infrastructure are very clear. Data center managers need agility, and you get that — connections are managed in software. And they need simpler connectivity. A virtualized infrastructure consolidates the connectivity hardware.
These benefits are going to attract a lot of different approaches, which is great for the buyer. But when you’re comparing solutions, it’s good to include a range of solutions, including Xsigo, in your analysis. Then you’ll know which delivers the best value for you.
Great observation. Xsigo’s server fabric is very much in the same camp as the software defined network guys. There are some interesting challenges in the SDN world. There appear to be two distinct approaches for SDN. One is openflow that requires every switch in the data path support the openflow API. Wile technically not all that difficult, it is not likely cisco and other leading switch vendors will jump on board because it effectively displaces the value add of their software stack such as cisco iOS and the newer NX-OS…..effectively this big expensive switches would become commodity packet forwarding devices under 3rd party control. This doesn’t appear to be happening beyond a few smaller players.
The biggest testament to this is the inventor of openflow has joined a startup pursuing another path. This alternative approach basically creates a private tunnel over ip ( specifically over udp or nvgre depending on which camp you belong to). Basically, this approach requires encapsulating Ethernet frames inside of an ip packet….which runs over Ethernet. The benefit of this is that the packets can traverse any network regardless of vendor. There are many challenges to this approach as well….including the security model, no tools to debug it, it requires software on the host tomprovide the encapsulation and de-encapsulation and it requires a separate gateway device to bridge between standard networks. Performance and security issues will certainly need to be resolved. Also, current models only work with hypervisors. Another issue is that in order to efficiently use this kind of technology, the underlying Ethernet network should be a fabric to efficiently handle multipathing. Lastly, these solutions were not designed to handle lossless storage traffic.
Xsigo’s solution leverages the maturity of the underlying infinibandd fabric. Infiniband has been around for about 12 years and scales easily to over 10,000 nodes with no manual configuration of the fabric…..it’s just plug and play. Infiniband also is industry leading in throughput, low latency, low power, and low cost compared to ethernet. Xsigo is not a technology bigot because they have an Ethernet fabric solution as well…..it just runs slower. Xsigo actually has a software defined network solution in it’s server fabric and it handles storage traffic as well as networking traffic. Xsigos io director is a gateway to bridge to to external networks, provides management and monitoring to guarantee standard security models hold true, and it works with any bare metal server or hypervisor.
Software defined networks ar here, they are real, they are proven, and deployed in the Xsigo server fabric.