– Matt Theurer, senior vice president, solutions architecture for Virtustream (www.virtustream.com), says:
Implementing a virtualization layer is now the norm. The high compute densities fostered by virtualization have placed similar stresses on the Local Area Network or LAN similar to the stresses placed on the SAN. Before virtualization the LAN traffic generated by a physical server was measured at a few hundred kilobits per second (Kbs) up to a few tens of megabits per second (Mbps). Modern physical hardware can have tens of processors processor cores and hundreds of gigabytes of RAM. This hardware supports tens and hundreds of virtual machines (vms) running on a single physical server with the concomitant network traffic now being measured in hundreds or even thousands of Mbps .
Fortunately, a new type of data center networks is emerging –converged networks – which shows great potential for handling scaled-up IT environments and streamlining data center operations.
Benefits of Convergence
Most traditional IT environments consist of two separate infrastructures – SAN and LAN – each with its own distinct architecture. This type of environment can be difficult to manage when scaling up and often requires two separate administrative groups to maintain. However, there is an alternative type of environment that is ideal for streamlining storage in data-intensive environments: a Converged Network Fabric. By merging SAN and LAN into one infrastructure, Converged Network Fabric merge can increase throughput, reduce cabling and redundant switches and adapters and lower administrative costs.
Because of their ability to simplify data center management, Converged Network Fabrics are ideal for enterprises considering migrating to a cloud computing environment. Data centers in traditional IT environments often have complicated communication networks, underutilized capacity and too many resources dedicated to managing both SAN and LAN infrastructures. But with a converged network, managers can strategically align SAN and LAN under one single administrative group, reducing management oversight and costs. In addition, converged network fabrics greatly reduce the number of physical connections per host. By reducing the physical infrastructure and simplifying data center management, converged network fabrics can help position enterprises for a smooth transition to the cloud.
Steps Toward Convergence
Prior to implementing a Converged Network Fabric it is important to understand the workload requirements of the existing communications and storage networks. You must assess the data transfer rates (typically measured in Kbps) for both storage and traditional network communications and you must understand the packets per second on the communications side and the input/output operations per second (IOPS) on the storage side.
Once you understand the storage and network traffic that will be crossing your converged fabric you must examine both the transit devices and the end point devices to verify that they can handle load. Your transit devices (switches) must be able to provide full point to point bandwidth as well as handle all of the packets traversing the interfaces. You will experience a larger mix of packets sizes from the traditional 1500 byte Ethernet frame to 2148 byte Fiber channel frames to 9000 byte Jumbo Ethernet frames. You absolutely do NOT want the ASICS in your switches to be oversubscribed. Beware of the inexpensive switch; it may cost you more in the long run.
Storage devices especially can put “back pressure” on the fabric and source load devices. By calculating IOPS (I/O per second) and the read/write ratio of those IOs, you can accurately design your backed storage system. Ensure your backend storage system(s) have sufficient I/O capacity to support the I/O being generated by the host(s). All of this data will help you determine the end benefit for your particular IT environment.
The most common mistake when implementing this type of convergence is not understanding the underlying traffic patterns. It is vitally important to size not only for bandwidth but also for packets per second on the communications side and IOPS on the storage side. Understanding how these types of traffic flow and interact is key to a successful implementation.
Another common mistake is not designing for high availability. On a converged fabric all data is flowing through a relatively small number of devices. Ensuring a sound, redundant design is critical.
The third most common issue is not sizing for growing workloads. This is especially true in virtualized and cloud infrastructures. Virtual server sprawl compounds this issue. Even in cases where the virtual infrastructure was properly sized and configured for network and disk I/O during the initial deployment, most organizations see dramatic growth in their virtual infrastructure over time. Typically additional network and storage resources are not added as new virtual machines are deployed. As new virtual machines appear, they are placed on existing disk pools and network volumes – often without consideration for the current communications and storage loads. Virtual machines use the same resources for a given workload as a physical server. It is important to have a good understanding of your requirements for both network and storage capacity and plan for growth accordingly.
Every day businesses are faced with massive data growth and flat IT budgets. Data centers in traditional IT environments often have complicated communication networks, underutilized capacity and high management costs. Converged networks, however, have the ability to increase data center efficiency and reduce management costs. By consolidating data center administration and streamlining throughput Converged Network Fabrics can help be a valuable solution for enterprises preparing for a move to the cloud.
For Additional Information
Two YouTube videos are available taking a closer look at Converged Fabrics:
Converged Fabrics: Part 2 – Calculating IOPs –