“To offload or not to offload?” That is the question that many 10Gb Ethernet (10GbE) Network Interface Card (NIC) vendors are asking IT managers to answer.

So here is the story…Intel and other NIC-only vendors are asking IT managers to run software initiators for iSCSI and Fibre Channel over Ethernet (FCoE). Their claim is that with all of the new CPU cores on next-generation servers, dedicating a few CPU cores to I/O is not a big deal and they will even tell you that it “makes sense.” However, let’s take a step back for a minute. Before server virtualization, I might have gone along with software initiators because a single application did not max out the CPU cores, so it did not matter. Then server virtualization came along and we started running three to four virtual machines (VMs) per CPU cores and this is when things began to change. Now, according to a presentation delivered by IDC in November 2008, we are moving to 10 to 12 VMs per CPU core. What is driving all of this? Data center consolidation.

The goal of maximizing the number of VMs has become a top priority for IT managers in their consolidation efforts and using CPU cores for I/O seems to be at odds with their key goals.

This is why Emulex vEngine™ CPU hardware offload technology, for both our OneConnect™ Universal Converged Network Adapters (UCNAs) and LightPulse® Fibre Channel Host Bus Adapters (HBAs), is such a benefit to IT managers. The ability of vEngine to fully offload I/O for IP, iSCSI and FCoE from the CPU has been shown to support up to 20% more VMs/CPU core. Now let’s extrapolate that to across 500 physical servers. If you are using FCoE or iSCSI software initiators, you are using 20% less for capital, power, cooling, cabling, switch ports and software licenses. Imagine if we could reduce 500 servers to 400 servers…what else could your IT group do with those capital expense (CAPEX) and operational expense (OPEX) resources? To learn more, take a look at our Convergenomics™ Calculator.

So, in addition to reducing data center consolidation, what are the compromises the open FCoE folks are asking IT managers to make? Here are the top five:

• CPU Efficiency and Virtualization – With FCoE software initiators, all of the Fibre Channel stack processing is done with server CPU resources. With OneConnect UCNAs, Fibre Channel stack processing is done by the adapter, which is especially critical for virtualized servers. Offloading (vEngine) Fibre Channel processing means more CPU resources can be used to support more VMs, providing greater cost savings and ROI for virtualized servers.
• Field-proven Enterprise Fibre Channel Stack – Emulex has spent more than 10 years, millions of man-hours and hundreds of millions of dollars developing, upgrading and testing Fibre Channel/FCoE firmware, drivers and APIs. This includes high-performance optimizations and, most importantly, robust error handling to achieve a proven and trusted solution. We have over 7 million installed Fibre Channel ports running in the world’s largest and most complex Storage Area Networks (SANs) across the globe, including more than 95 of the top 100 data centers. Emulex FCoE UCNAs are the field-proven, no-risk solution. Unproven FCoE implementations from network interface card (NIC) vendors present high-risk and uncertainty, which is unacceptable for enterprise-class data centers.
• Enterprise-class Reliability and Support – Customers who buy Emulex OneConnect FCoE CNAs can rely on a rock-solid line of support that includes leading server and storage partners, value-added resellers (VARs) and distributors, all backed by Emulex. This multi-tier level of support means customers know that help is available for simple questions and hard-to-diagnose problems from experienced Fibre Channel experts. This level of support took years to develop and mature. Open source and first-release solutions have a much different paradigm. If we use open source Linux as an example, it was only moved to enterprise applications when leading vendors provided a stable kernel, ecosystem testing and proven interoperability. It will take years to reach that level of support for FCoE software initiators.
• Enterprise-class Interoperability – These benefits are only realized if FCoE solutions are fully tested and interoperable with existing Fibre Channel investments, which could go back many years and cover many product generations. That is only possible with a Fibre Channel vendor like Emulex, which continues full support and qualification for 2, 4 and 8Gb Fibre Channel target devices. Emulex CNAs use the same HBA software stack and management console that has been qualified by all of the leading server and storage vendors.
• Operating System Qualifications – Emulex drivers have been certified for Microsoft Windows, Solaris, AIX, HP-UX, Red Hat, SuSE, VMware, Hyper-V, Citrix Xen and more. To support this effort, Emulex has dedicated engineering teams for each operating system (OS) and hypervisor. Based on a long-standing relationship with the Fibre Channel ecosystem, Emulex drivers and management software and APIs are fully tested and used to integrate with a wide range of OS features and management models. This is built on ongoing, fully engaged relationships with OS and application development groups that work on storage.

New FCoE software initiators and partial offloads have a much different story. They have software stacks and management applications that will need to go through qualification and release cycles that could take several years to complete. APIs to support OS and application integration are either immature or non-existent.

Open source solutions can be great for many applications, but when it comes to enterprise-class FCoE, the open FCoE driver set not only fails to meet the need of IT, they are detrimental to their goals of data center consolidation.

Posted in Uncategorized | 1 Comment