Most folks are familiar with virtual machines. Virtual machines provide a complete simulation of actual underlying hardware. Software like VMware allows data center managers to run one or many virtual machines on a single server. VMware software running on a server host reads a (usually large) data file that represents the contents of a "hard dive" that contains the guest operating system, applications, and data. Virtual machines can be moved from server to server as their need for greater processing power increases. Virtual machines are normally operating system independent in the sense that a virtual machine can run as a guest on any operating system supported by the virtual machine software vendor. Virtual machines have prompted data center managers to undertake "server consolidation" activities.
Some virtual machine software implementations include: VMware Workstation, VMware Server (formerly GSX Server), VirtualBox, Parallels Desktop, Adeos, Mac-on-Linux, Win4BSD, and Win4Lin Pro.
What about "I/O Virtualization" or "Virtual I/O?"
Xsigo Systems, Inc. aims to define I/O Virtualization the way the VMware has defined server virtualization. Xsigo emerged from three years of secrecy at the VMWorld conference on September 12, 2007 announcing the company and the technology. The company is backed by Kleiner Perkins,
Khosla Ventures, and Greylock Partners. Read more about the announcement. The idea is to make cable changes and application moves simpler.
Here's the Xsigo pitch: "The faster, simpler, more cost effective way to connect servers to networks and storage.
- 70% fewer cables
- Up to 50% less capital cost
- 80% simpler I/O management
- Open systems interoperability"
It appears that they are selling a box called the
Xsigo I/O Director, interface cards, and software drivers. The box occupies 4u in a rack and can be configured with multiple modules: 4 x 1Gb Ethernet modules, 1 x 10Gb Ethernet modules, 2 x 4Gb Fibre Channel modules, and/or SSL modules. That last item may require a little explanation. Quoted: "Consolidate SSL functionality from multiple servers to just one card. This option permits secure communications to and from all attached servers and removes the cost and complexity of server-based SSL."
The box can be managed through a command line interpreter (CLI) or its graphical user interface (GUI). You make connectivity changes through drag and drop operations. The I/O Director also integrates with existing management frameworks through an open API.
In practice, you would install an interface card from Xsigo into each server. Quoted: "A single host channel adapter (HCA) in each server consolidates storage and network connectivity. Each HCA can deploy up to 32 virtual host bus adapters (HBAs) and 32 virtual network interface cards NICs within the server. Other than a standard driver, no agents or management software are installed on the server." The idea is that you can deploy virtual NICs and HBAs on the fly to create application-specific connectivity. Virtual resources appear exactly as their physical counterparts. The switch fabric is non-blocking at 10Gbps.
Once you have all this set up, you can do things like:
- assign applications on a server to their own IP address (via a virtual interface card)
- move an application from one server to another (and keep the same ip address)
- move a virtual machine from one server to another and keep the same IP address
- ensure that virtual machines running on the same server get the priority they should because the Xsigo software can prioritize the I/O to each virtual I/O port. This could be used in QOS agreements.
- one server can quickly adopt the identity of another, making fail overs rapid.
- duplicate configurations at failover sites.
- speed up backup jobs because the network speed is no longer a limiting factor.
Xsigo's website gives this contact info:
Xsigo Systems; 940 Stewart Drive; Sunnyvale, CA 94085 Tel: 408-329-5600. Mysteriously, a brochure give this contact info: Headquarters: Xsigo Systems, Inc.; 70 West Plumeria Drive; San Jose, CA 95134 USA Tel: 408-329-5600 www.xsigo.com. Maybe they just moved.
Analysis: Let's believe for a moment that all this works and there are no bugs in this new system. Let's also grant that this startup has credibility with seasoned management and technical expertise. Who's going to buy this? Really that's the same question as "who has the wherewithal to experiment and take the time to help this new company eliminate the bugs in this new system?" A few names come to mind: Citigroup, Wells Fargo, IBM, AT&T, Google, maybe a large datacenter operation somewhere. The venture firms backing Xsigo have deep pockets. Still you have to wonder what they were thinking. Is there sufficient pain in datacenters to warrant the trouble involved in setting up an Xsigo I/O Director, the interface cards and the drivers? New technologies prompt management to upgrade hardware on a regular basis, so how are they coping right now? What does Google do? There are problems that the Xsigo does not solve. So what percent of all problems are solved by Xsigo? Will it be possible to eliminate one position thereby justifying the cost of the Xsigo system? For some datacenters maybe. Enough to keep this company going? The venture backers must have figured that this company would make a good buyout candidate for Cisco, Juniper, HP, Dell, or even IBM. Operating in the context of a larger product offering with a sales pitch of reducing costs makes sense.
We'll examine competition in a future post.
