The Case For I/O Virtualization

When the term "virtualization" is used, the implicit assumption is that we're talking about "server virtualization." But other types of virtualization exist, such as storage virtualization. And now I/O (input/output) virtualization is starting to encroach upon our consciousness.

Just what is I/O virtualization?  Answering that question requires a brief review of the role of I/O in a system. The job of a computer is to perform "data processing." A CPU processes data from its associated transient memory, but the data typically has to be acquired from storage somewhere, and a network is used to link the two. The movement of data is all about I/Os (inputs are reads from storage to a server and outputs are writes to storage).

To be clear, the use of network may be grandiose when using direct attached storage where a bus or I/O interconnect is sufficient rather than a LAN or SAN, but logically any connectivity is still part of a network.

Now if I/O needs to get in or out of the server chassis box in conjunction with an external network composed of such things as cables and switches, an I/O adapter is necessary to connect and translate network protocols to the CPU/memory complex of a server. If the network is a SAN, a Fibre Channel HBA (host bus adapter) may be the I/O adapter. Or, if the connection is to a LAN or WAN, an Ethernet NIC (network interface card) may be used. However, the I/O adapter is located within the server box.

What I/O virtualization does is move these adapters from the server box to a new box called an I/O virtualization switch box that is external to the server. VirtenSys is joining the I/O virtualization party (where NextIO has played for over a year) by providing such a switch. With the VirtenSys solution, a single PCIe cable connection is made between each physical server and the I/O virtualization switch. Each server no longer has a dedicated I/O adapter, but rather has virtual I/O connectivity through non-dedicated access to I/O adapter resources on the I/O virtualization switch.What does this do and why does it matter? In general, one of the big drivers of virtualization has been that IT assets are underutilized, a situation that has been true of I/O adapters. VirtenSys has found that adapters are only utilized 10% to 15% of the time, and their virtualization switch can increase that to more than 80%.

Using VirtenSys' solutions, an IT organization can defer the need to purchase additional I/O adapters until the demand for I/O increases sufficiently to warrant them. And even when additional purchases are needed, fewer adapters will be required than in the past. Moreover, some currently redundant I/O adapters can simply be taken out of service. That reduces the power and concomitant cooling requirements. That is nothing to sneeze at as up to 30% of power is consumed by I/O cards and switches.

Those are obvious benefits, but there are a couple of other points that, although not as obvious, can be critical. However, we must also note that a VirtenSys switch is non-disruptive in that it works with the existing infrastructure in data centers. That means that no changes to adapters and no non-standard hardware or drivers are required, as well as no change to existing drivers, software and operating system infrastructures. In addition, the performance transparency of VirtenSys's switch means there is zero to minimal performance overhead. So except for inserting a new switch box in the network, no real barriers to adoption exist.

But that transparency leads to a hidden benefit. The VirtenSys switch creates a virtual I/O cloud that can accommodate different bandwidth requirements. As Ethernet, Fibre Channel, and SAS increase their bandwidths over time, the old and the new bandwidths can be mixed and matched so transitions can be accommodated incrementally rather than as a rip and replace. IT can move in an evolutionary manner rather than a revolutionary manner, meaning that organizations can buy bandwidth upgrades when they can afford them and phase in new implementations when IT staff and the infrastructure are ready.

The second benefit -- that is that I/O virtualization enables denser server configurations -- may be a little more obvious. Removing the I/O from a server leaves only a CPU and memory, so what was once a 2U configuration can often fit into a 1U rack, opening up critical space.Another side effect of virtualizing I/O is that the number of cables can be dramatically decreased. Many a data center managers would gladly allow an extra switch box just for that benefit alone, as cable problems are notoriously hard to diagnose. For that reason, I/O virtualization is a good complement to blade environments as it simplifies all the processes required to manage blades, such as configuring, failing over and load balancing.

The first reaction that a data center manager might have to I/O virtualization is "Not another virtualization story. Not another box I have to put in my network." However, upon further reflection, that same data center manager will probably see that I/O virtualization is eminently sensible. If the transformation of a data center is going to take place anyway, whether that is called a next generation data center, a utility, or an internal cloud, half measures are not going to be enough. All forms of virtualization should be examined and adopted as appropriate.

A good case can be made that I/O virtualization should be part of the package. The any-to-any connectivity that I/O virtualization offers makes traditional fixed I/O infrastructures essentially obsolete. The VirtenSys solution makes true I/O resource sharing possible at the same time as it ensures complete I/O compatibility. Overall, I believe that carefully examining VirtenSys' solutions is a good move for companies considering I/O virtualization or datacenter transformation.