Broadcom Doubles VMware CPU Efficiency
A few weeks ago, VMware announced that vSphere 4.1 enables 10Gb iSCSI hardware offload in the Broadcom 957711 Converged NIC (C-NIC). Before VMware supported 10Gb iSCSI offload, data center managers connected their VMware servers to 10Gb iSCSI SANs using a combination of NIC hardware and fat iSCSI device drivers, a configuration where most of the iSCSI protocol processing is done by the VMware server. Now with vSphere 4.1, VMware supports the latest generation of converged C-NICs that use thin iS
September 24, 2010
A few weeks ago, VMware announced that vSphere 4.1 enables 10Gb iSCSI hardware offload in the Broadcom 957711 Converged NIC (C-NIC). Before VMware supported 10Gb iSCSI offload, data center managers connected their VMware servers to 10Gb iSCSI SANs using a combination of NIC hardware and fat iSCSI device drivers, a configuration where most of the iSCSI protocol processing is done by the VMware server. Now with vSphere 4.1, VMware supports the latest generation of converged C-NICs that use thin iSCSI drivers, a scenario where most of the iSCSI protocol processing is done by the iSCSI offload engine in the C-NIC. My expectation for 10Gb iSCSI offload is that storage deployed with VSphere 4.1 will run faster and use a lot less of the precious CPU resources that rightfully belong to virtual machines and business applications.
According to Page Tagizad, product line director for Broadcom High Speed Controllers, "the BCM957711 10Gb C-NIC is the exclusive 10Gb iSCSI offload qualified by VMware."
Last week, Broadcom published iSCSI offload test results that exceeded my expectations for IOPS performance (input/ output operations per second) and CPU efficiency (IOPS per percent of CPU utilization). Broadcom demonstrated that iSCSI HBA offload doubles the performance of VMware vSphere I/O in the application performance sweet spots of 4KB and 8KB block sizes. At the 4KB block size commonly used with Microsoft Exchange deployments, the use of iSCSI HBA almost triples performance. At the 8KB block size commonly used with Oracle online transaction processing (OLTP) deployments, the use of iSCSI HBA more than doubles IOPS performance.
Testing by Broadcom also shows that iSCSI offload technology doubles the CPU efficiency of servers running VMware vSphere I/O with iSCSI storage. At the 4KB block size used with Microsoft Exchange, the use of iSCSI offload almost triples CPU effectiveness. At the 8KB block size used with Oracle OLTP, the use of iSCSI again almost triples CPU efficiency.
In a recent IT Brand Pulse survey, data center managers said that server virtualization was the second most important operating system in their data center next to Microsoft Windows. Given the critical importance of server virtualization and the fact that CPU efficiency translates into more VMs per server. I consider the dramatic increase in CPU efficiency to be a breakthrough achievement by Broadcom. Impressively, by more than doubling CPU efficiency, Broadcom instantly more than doubled the number of VMs that can be supported per server. With vSphere 4.1, a new best practice has been established for server virtualization and iSCSI storage: iSCSI offload must be enabled. If you are a data center administrator responsible for virtual server connectivity to an iSCSI SAN, iSCSI offload is a "must have" feature in your 10Gb Converged NIC.
About the Author
You May Also Like