Performance Considerations in Storage Virtualization

Storage virtualization technology not only means storing the storage pool to the host, but also ensuring that the data passes through the virtual controller without impacting performance, so many users ask what is the IOPS value of the virtual storage vendor product?

This is not an easy question to answer, at least for today there are no third-party agencies, such as SPC, that provide IOPS measurements for the overall solution of storage virtualization. The performance parameter values provided by each storage storage vendor are often not referenced, such as the ability to stack the front-end and back-end throughput of the virtual controller, regardless of the processing performance of the actual controllers, which can lead to serious misdirection of the user- Especially when the user uses the purchasing experience of the disk array to purchase the storage virtualization solution, it applies the previous experience directly.

There is no IOPS, the concept of IO per second, for storage virtualization products or scenarios. Most virtualization products can increase data throughput by extending the IO card in parallel or on a single node, however, due to the different communication strategies of the back-end disk array, the virtualization of the product on the virtualization scheduling of the additional performance overhead, that is, after the use of virtualization products, without adding additional high-performance disk on the premise of , the throughput consolidation of the original disk array must be less than the data throughput of the virtualization product. Therefore, in assessing the existing systems to be consolidated, it should be fully realized that not all business systems fit together for virtualization, such as two or more applications that peak performance at the same time should be avoided in the same storage pool, and for core applications that have been heavily loaded, The original system architecture of isolated storage should still be used.

Depending on the level of deployment, the storage virtualization solution is divided into the host layer, the swap layer, and the storage layer implementation. Storage virtualization solutions at the host layer enable storage virtualization by installing virtualization software on the host side its performance overhead is mainly generated on the host side; the performance bottleneck of the exchange layer is mainly on the virtualization controller, such as IBMSVC, the maximum extensibility of products is generally around 4, when there is a large amount of storage space in the background need to be managed, And the use of disaster-tolerant or storage tiering will greatly increase the burden of the virtualization controller; Disk array virtualization is also at risk for performance bottlenecks, but it can boost processing capabilities with the help of disk array internal controllers. Furthermore, the virtualization of the storage layer enables the performance of the disk arrays attached to it to be elevated, unlike the previous dynamic layering concept.

After storage virtualization, the structure of the disk in the background is complex and requires high-performance allocation of storage pools to critical applications based on application-level policies, so policy-based performance management is the most important part of storage virtualization. Currently there are two types of technologies in storage performance management, one is traditional dynamic storage rating, the other is policy-based storage acceleration technology. Traditional dynamic tiered storage can only adjust the application of busy I/O from slow disk to the high-speed disk system, improving performance in this way can partially improve performance improvements, but this pattern is at the expense of the storage virtualization Gateway's own resources, while increasing the frequent scheduling between storage I/O. Overhead storage resources, which have a negative impact on the overall performance of the storage, and have limited performance improvements, as well as uncertainties about how the storage space is distributed. In a cloud computing environment, more performance-accelerated technologies for application policies are used as storage virtualization performance management, which extends the concept of caching and makes storage virtualization a flexible, simple solution that requires high background performance.