Welcome!

Virtualization Authors: Rebel Brown, Miko Matsumura, Yeshim Deniz, Liz McMillan, Chris Fleck

Related Topics: Virtualization

Virtualization: Article

I/O Virtualization: Scalability and Performance

Implementing the vision of stateless computing

There are pitfalls involved in using software-based network interface cards (NIC) and host bus adaptors (HBA) as they don’t scale well in I/O-centric loads running on scaled-out architectures. The CEO of one of the leading virtualization companies has said, “Software virtualization addresses 80% of the market and it leaves 20% for hardware-based I/O virtualization.” Scaled-out architectures and stateless computing are the norm in the computing business. I/O devices are being pulled out and connected through PCI Express Gen 2 as a step toward scaled-out architectures. A second step is to combine Ethernet and Fibre Channel onto Fiber Channel Over Ethernet. FCoE scales well and provides unparalleled performance to the system. The number of I/O connections and I/O devices required per server will be limited to the number required at the bottom of the PCIe aggregation layer. This architecture and usage model will change the fundamental economics, power and cooling of next-generation data centers. Along with I/O virtualization and network convergence, I/O virtualization will significantly reduce TCO and OPEX.

Traditional Architecture
Traditional architectures have servers, NICs, and HBAs feeding into access layer switches. These switches feed into edge switches or core/director class switches in the data centers (see Figure 1). The application appliances are on the other side of the core of the network. This architecture scales well but does not solve the server underutilization problem. To address that, stateless computing architectures are being implemented that view the computer as being a CPU complex plus memory and abstract the I/O. This in turn allows on-the-fly migration of I/O devices, application migration, and autonomic computing, which allows any application to run on any server. With these capabilities, multiple applications can run on a single physical server, resulting in high server utilization and lower overall costs.

Virtualization
To increase the performance and utilization, multiple virtual machines are being run on the servers. The hypervisor resides in a new software layer between the OS and the hardware and provides the ability to run multiple virtual servers on a single machine. This capability can be extended by I/O virtualization and network convergence in the hardware. The hardware needs to provide security and a quality of service (QoS) that ensures the QoS on each virtual resource. The virtual NICs and virtual HBAs are implemented in hypervisors as shown in Figure 2.

About Shreyas Shah

Shreyas Shah is chief systems architect at PLX Technology, responsible for architecture and system applications of PLX's PCI Express IOV switching products. He has nearly 15+ years of experience in engineering, with an emphasis on the computing, networking and storage communications market segment. Before joining PLX, he held engineering positions in chip and system architecture at Alantec/FORE/Marconi communications, Sisilk networks, Fabric7 Systems. Lastly he was Founder and CTO of Xsigo systems (IO Virtualization Company). Shreyas holds an MSEE from the Indian Institute of Technology, India.

Comments (0)

Share your thoughts on this story.

Add your comment
You must be signed in to add a comment. Sign-in | Register

In accordance with our Comment Policy, we encourage comments that are on topic, relevant and to-the-point. We will remove comments that include profanity, personal attacks, racial slurs, threats of violence, or other inappropriate material that violates our Terms and Conditions, and will block users who make repeated violations. We ask all readers to expect diversity of opinion and to treat one another with dignity and respect.