The 8Gb/s Fibre Channel transition is different from the 1Gb/s to 2Gb/s and 2Gb/s to 4Gb/s transitions. The major reason why? It’s not free. The cost of optics in the 8Gb/s transition changed this predictable and smooth progression. Additionally, in each of the previous generations, the key drivers of the speed transitions were RAID (redundant array of independent disks) and tape target devices and the introduction of PCI Express. In contrast, for the 8Gb/s Fibre Channel transition, the drivers are from the host side, and this is due to a new technology trifecta…virtualization, blades and the new Intel® Xeon® processor 5500 series (formerly codenamed Nehalem). These three technologies are driving and enabling greater I/O demands per system and making now the right time to move to 8Gb/s Fibre Channel.
I/O Aggregation with Virtualization
One of the realities of virtualization is that the I/O workload of many systems is aggregated onto a single or fewer systems. This drives the need for two key storage functions, networking for virtual machine (VM) mobility provided by Storage Area Networks (SANs) and performance bandwidth to handle the consolidated workloads. A recent IDC research report indicated the number of VMs per server will grow from three to four per computer processing unit (CPU) to 10 to 12 per CPU. This growth in I/O workload on a single system will require the performance of 8Gb/s Fibre Channel Host Bus Adapters (HBAs) or 10Gb/s Ethernet Converged Network Adapters (CNAs).
Blade Connectivity Limitations
Blades are great for consolidation, environmental savings and maximizing computing density. However, they do force IT managers to be very budget conscious when it comes to their I/O connectivity choices. The limited connectivity on blade servers will quickly require the move to 8Gb/s Fibre Channel to meet the increased I/O workloads.
Intel Maximizes Virtualization and I/O Performance with its Xeon® Processor 5500 Series (Nehalem)
The Intel Xeon 5500 processor series has a number of key advancements that Emulex’s LightPulse® architecture maximizes, including:
- Intel QuickPath Interface (QPI) – Provides high-speed connections between the CPU cores, external memory and the I/O hub. This enables the same 8Gb/s Fibre Channel HBA to deliver 225% better performance vs. previous generations of Intel processors.
- Message Signaled Interrupts eXtended (MSI-X) – Allows CPU interrupts from multiple devices to be processed in parallel, eliminating idle time spent waiting for individual interrupts to complete. This enables Emulex to scale I/O to support more VMs per CPU.
- PCI Express 2.0 – The Intel Xeon processor 5500 series and Emulex 8Gb/s HBAs both support PCI Express 2.0, which doubles the internal bus speed for maximum performance.
- Intel Hyper-Threading Technology (HTT) – Hyper-threading allows each CPU to be divided into two threads, providing more parallel processing and better performance for multi-threaded applications. For Emulex adapters, hyper-threading doubles the number of concurrent I/O and interrupt threads, maximizing the benefit of MSI-X and NUMA support.
- Virtualization Technology for Directed I/O (VT-d) – The Intel Xeon 5500 processor provides complete support for Intel Virtualization Technology, which includes VT-d for improved I/O performance. VT-d support has also been added to the high-performance Emulex LightPulse 8Gb/s Fibre Channel HBA family.
Dell’Oro: Documenting the Shift to Emulex’s 8Gb
According to the Dell’Oro Group, Emulex is driving greater share in the 8Gb/s Fibre Channel transition. Our strong execution on 8Gb/s Fibre Channel qualifications, vEngine I/O offload architecture, leading position with server virtualization, performance, scalability and proven enterprise reliability are enabling Emulex to gain share in the 8Gb/s Fibre Channel market transition.
Positioning for the Future
Beginning the transition to 8Gb/s Fibre Channel is a smart move for customers and partners, it will enable them to capitalize on the value of their new Intel Xeon 5500 processor-based system investments, maximize their consolidation with blades and provide the most scalable I/O solution for virtualization platforms during the next CPU cycle.
Posted in Virtualization | No Comments