Implementing SR-IOV on HP ProLiant Servers with VMware vSphere 5.1

Technical white paper | Implementing SR-IOV on HP ProLiant Servers with VMware vSphere 5.1
2
Introduction
HP and VMware enable Single Root I/O Virtualization (SR-IOV) for VMware vSphere 5.1 on HP ProLiant Servers. SR-IOV
enables a single PCI Express (PCIe) network adapter to appear to the VMware vSphere (ESXi) hypervisor as multiple
special-purpose network adapters. These special-purpose network adapters, termed Virtual Functions (VF), are only
available for direct presentation to virtual machines (VMs). By providing a VF directly to a VM, the hypervisor’s virtual switch
(vSwitch) is no longer required to process network traffic. This hypervisor bypass increases network throughput, lowers
latency, and reduces overall CPU utilization.
Prior to vSphere 5.1 hypervisor bypass was restricted to VMDirectPath. VMDirectPath requires that an entire PCI function be
provided to each VM. In the case of a network adapter, this resulted in one physical network port being dedicated to a single
VM. While this may be appropriate for some situations, VMDirectPath does not scale well. With SR-IOV, one physical network
port can be shared by multiple VMs while still providing hypervisor bypass.
This paper focuses on the requirements for a successful SR-IOV implementation with vSphere 5.1 on ProLiant servers.
When planning an SR-IOV implementation, take into account the configuration and support restrictions discussed in this
paper. Additionally, this paper assumes that you are familiar with vSphere and ProLiant servers. For additional information
on these products, see hp.com/go/proliant
and hp.com/go/vmware.
SR-IOV Overview
Network adapters that feature SR-IOV are comprised of one Physical Function (PF) and multiple VFs per port. The PF is
responsible for the management and configuration of its associated VFs. On the host server, the VMware ESXi administrator
configures a PF to present a defined number of VFs. This process is called enumerating the VFs. During VM configuration,
the enumerated VFs are available for selection as a PCI device. To the Guest OS, the VF appears as a single-port network
adapter. Each VM must use a unique VF; sharing a VF between two or more VMs is restricted.