But it deserves a fanfare because it gives KVM the ability to handle unprecedented storage I/O rates more than 50% higher than its competition. This is important for running the most demanding mission-critical workloads—like databases, ERP, and low-latency financial trading applications —because previous technologies weren’t fast enough to process them competitively in virtual or cloud environments.
This has changed with the SP3 Technology Preview and enterprise-class hardware. Such applications can now run on VMs and clouds, giving you benefits like physical server consolidation and greater energy efficiency for substantial cost savings.
This advance is the result of the SUSE/IBM relationship and SUSE’s virtualization strategy. SUSE and IBM have a 13-year partnership of joint development. For example, IBM-SUSE teams test, lock down, and polish all SUSE betas and Linux releases behind the scenes. As of SP3 Launch, 423 new OS Certification Bulletins for IBM servers have already been posted. And SUSE has a cross-virtualization platform approach, which means our products (e.g., SUSE Studio, SUSE Manager, SUSE Cloud) are designed to ensure optimal performance on all major open and commercial hypervisors, including KVM.
While IBM is well-known as a KVM champion, SUSE may not spring to mind when you think of KVM. We don’t sell it as a separate “product.” Rather, KVM is part our core technology. Along with Xen, KVM is included in the virtualization subscription for SUSE Linux Enterprise Server 11 SP3 and for SUSE Linux Enterprise Server for System z 11 SP3. KVM support is also fundamental to our other products: SUSE Cloud, our OpenStack-based IaaS platform for deploying and managing private clouds; SUSE Studio, a tool for building images for physical, virtual, and cloud environments; and SUSE Manager, software for managing SUSE Linux Enterprise Server across major hardware architectures, virtual platforms and cloud environments.
All of this optimized technology sets the stage for the new virtio-blk-data-plane technology and its test in the IBM Linux Technology Center. This new I/O virtualization technology accelerates operations going through the para-virtualized I/O block driver (virtio-blk) with dedicated per-device threads. This allows the block I/O processing to run concurrently with other device emulation to achieve super-fast I/O performance results.
The test setup included an IBM System x3850 X5 host server with QLogic QLE 256x Host Bus Adapters and SUSE Linux Enterprise Server 11 SP3—with virtio-blk-data-plane technology in its Technology Preview—running on both the host server and on a single KVM guest. You can read the whole test story in “KVM Virtualized I/O Performance.”
Direct Random I/Os Across Various Block Size
The test setup produced the highest storage performance ever recorded for a virtual machine: 1.37 million I/O operations per second (IOPS) at 8KB I/O request size and more than 1.61 million IOPS at 4KB and smaller request sizes. Now consider this: Microsoft benchmark performance results for iSCSI using its Hyper-V hypervisor, Windows Server 2008 R2, and Intel Internet Server Adapters achieved an aggregate rate of 700,000 IOPS using 10 virtual machines on a single host at 512-byte I/O request size. And VMware reported almost 1.1 million IOPS using 4KB I/O requests on a single virtual machine running on a vSphere 5.1 host.
In other words, the rates produced by the SP3 Technology Preview running on KVM produced more than 50% higher rates. That’s a breakthrough that you really need to read more about!
|Add Comment (please login)|