Flash Storage Doubles VMware vCloud Performance
Server virtualization has been credited for improving server resource utilization. Gartner estimates that 75% of all x86 servers are virtualized. The next step of virtualization is the ability to deliver IT as a service. Many organizations are turning to VMware vCloud Suite to accomplish this.
VMware vCloud Suite bundles several components including vSphere, vCenter SRM, vRealize Operations, vRealize Automation, and vRealize business to improve virtualization, automating IT service delivery, and bring applications to market faster. While virtualization can provide utilization benefits, it can also create some challenges. One of those challenges is I/O access to storage.
Storage Management Improvement
Virtualized workloads randomly try to access data. It is no longer a linear, predictable process. VMware tries to manage this, but they realized that some of that could be better offloaded to the storage device. At the same time, vSphere could function better if it understood the underlying data storage. To accomplish this, VMware developed a set of APIs to enable collaboration between vCloud and the storage device.
Those APIs are:
- VAAI = VMware vStorage APIs for Array Integration
- VASA = VMware vSphere APIs for Storage Awareness
VAAI are the APIs that VMware uses to send certain functions to the storage array. VASA, on the other hand, are the APIs that allow the storage array to send information about itself to vCloud. Together, these tools work together to best optimize performance, offload CPU functions, and automate storage optimization based on the workload, data, and attributes of the hardware.
VMware Partnership and Integration
As with any APIs that a software vendor develops, it is up to the hardware manufacturer to take advantage of them. How those manufacturers adopt, optimize, and maximize the benefits of the integration will determine the particular performance gains that can be realized.
IBM saw the opportunity to develop an incredible value proposition by taking full advantage of this opportunity. They partnered closely with VMware and worked together for years to bring that value to IBM FlashSystem V9000.
There are several key system components that are now built in FlashSystem V9000. Those are hardware-accelerated Block Zero, Full Copy, and hardware-assisted Locking. FlashSystem also provides as much data as possible back to vCloud through the VASA APIs. All of these systems are enabled by default, but they can also be tuned, modified, and adjusted by the administrator.
vCloud and IBM Flash Speed Test
IBM and VMware wanted to validate the performance gains that customers were reporting. So the two tech giants partnered to test the system at the IBM Flash Center of Competency in Gaithersburg, Maryland. The performance results were extremely impressive. No one has the same storage environment, so some organizations would experience different results.
In the benchmark study, they ran vCloud on IBM 3650 x86 servers connected to IBM FlashSystem V9000 with 20TB of usable storage. For the comparison, they set up enterprise disk storage consisting of 24 300GB 15K RPM drives with multiple RAIDs. On the system, they ran HammerDB.
HammerDB is an open source database load testing and benchmarking tool for Oracle, SQL Server, DB2, TimesTen, MySQL, MariaDB, PostgreSQL, and Postgres.
In the virtual environment, the systems running on enterprise hard disk drives were capable of 150,000transactions per minute (TPM). Then they ran the same workloads on the system connected to FlashSystem. The performance more than doubled to almost 400,000 TPMs. The performance on flash also did not significantly drop off when compression was enabled.
IBM has demonstrated that an organization could achieve huge performance gains while also reducing the storage footprint with compression. Flash technology has been expanding in the data center because of its performance capabilities. When that flash storage is tuned to work with a specific workload such as vCloud, the results can even be more impressive.