Enterprise Strategy Group
Published: 13 Mar 2008
| I/O obstacles could throw a monkey wrench into virtual server environments, but there are alternatives to traditional methods.
To take full advantage of server virtualization benefits like app mobility and high-availability/disaster recovery (HA/DR) capabilities, virtualized servers need networked storage. According to a recent ESG research report, "Server Virtualization: The Impact on Storage," 86% of respondents using server virtualization support their virtualized environments with networked storage, and the majority of those are Fibre Channel (FC) SANs. The result is a highly dynamic infrastructure in which every server device requires connectivity to every storage device to ensure flexibility and mobility. However, the I/O path is still being provisioned as it would be in a fixed, nonvirtualized environment. To keep pace with changes in the server and storage domains, organizations need to ensure that the I/O path is not a bottleneck.
One of the most obvious issues with this type of arrangement is the cost associated with deploying and maintaining an environment that, in a high-availability configuration, will sport two NIC cards and two HBAs per server, in addition to all of the cabling that goes with it.
Troubleshooting an environment with multiple connections becomes difficult because each path may be a point of failure. Some of the most difficult problems to track are related to cables. It's not uncommon for IT staff to spend days trying to troubleshoot a problem only to find that it can be traced to a cable with an extreme bend in it.
There's also the complexity inherent in provisioning new physical servers in enterprises with separate networking and storage groups, both of which need to coordinate to provision NIC and HBA cards. In some cases, this could take weeks.
Traditional HBAs can't see virtual machines, which creates potential security and management issues. From a security standpoint, all applications running on a physical server must be in the same zone. From a management standpoint, it's difficult to tie applications to the storage if you can't identify the virtual machines. Ultimately, you need a better solution for greater scale and flexibility, and improved functionality.
There are solutions that address these problems, but the extent to which they solve them varies. Let's take a look at some of the technologies looking to resolve these problems.
Automated patch panels. This option offers the ability to wire all connections into a single patch panel, which can then handle all reconfigurations. This solution has been available for many years and is very popular in lab configurations. (However, it doesn't address the need to see virtual machines.) For those environments with constantly changing physical environments, this would be very useful. Key suppliers in this market include Apcon and OnPath Technologies (formerly IntelliPath).
N_Port ID Virtualization (NPIV). NPIV recognizes multiple IDs from one physical HBA, essentially creating logical HBA ports. This technology was specifically designed to recognize virtual machines on a physical server. Many refer to it as virtual HBAs. The technology is focused mainly on FC, so while it doesn't reduce the number of cables, it does provide additional intelligence in virtualized servers and reduces some security and management burdens. Key suppliers of this technology are Emulex and QLogic.
Fibre Channel over Ethernet (FCoE). Here's another technology looking to solve the I/O problem. Based on developing enhancements to existing Ethernet protocols like priority flow control, lossless Ethernet, congestion management and priority grouping, FCoE will reduce connectivity requirements. It does this by using a single converged network adapter to handle both IP and FC traffic through a single physical interface. It will support 10Gb throughput and current FC management solutions. This technology is still emerging and products (demonstrated at trade shows) aren't expected to be widely available until late 2008 or early 2009. There's a great deal of support for this technology from many of the industry's leading providers like Brocade, Cisco, Emulex, Network Appliance, Nuova and QLogic.
InfiniBand. This option provides high throughput and is currently available. Companies are taking advantage of this technology to enable I/O virtualization; essentially, a single card can have multiple virtual NICs or virtual HBAs. This type of solution eliminates the need for multiple cables, NICs and HBAs, and provides sufficient bandwidth to eliminate any I/O bottlenecks. One company pioneering virtual I/O with InfiniBand is Xsigo Systems. The company has solutions available today, and has also built an open solution that will enable it to convert to FCoE when it's available.
Implementing technology for the sake of technology doesn't make sense. Deploying new technology because it saves you time and money, and delivers higher levels of service, is worth investigating. The benefits of virtual I/O include:
Virtualizing the I/O path may be the best way to accommodate the dynamic nature of these environments. There's a variety of technologies available--with more on the horizon--that address some or all of the issues created by server virtualization. The first step is to become familiar with the various techniques and to decide what makes the most sense for you today and in the future.