Why the I/O blender causes storage latency in virtual environmentsDate: Aug 29, 2013
In this video, DeepStorage.net founder Howard Marks sat down with Storage Media Group news and features writer Todd Erickson to discuss the I/O blender effect in virtual environments.
Random I/O is a source of storage latency created when a computing environment switches from physical servers to virtual servers. The problem is commonly referred to as the I/O blender effect.
In traditional physical server storage environments, servers are dedicated to storage systems, and the storage systems can predict where a request is located before serving that request. However, virtual servers force multiple workloads to communicate with the same volume simultaneously, therefore mixing up -- or blending -- the I/O requests.
According to Marks, this means storage in a virtual environment can't take advantage of features like pre-fetching, and therefore "needs a bit more performance from the storage system than that same set of workloads in the physical world."
Unfortunately, the I/O blender is an aspect of virtual environments that can't be completely eliminated, so virtualization and storage administrators need to find ways to reduce the resulting storage latency. Perhaps the most common way to mitigate the I/O blender effect is to use solid-state drives (SSDs) for cache. The high-performance nature of SSDs translates to less latency overall.
As the number of virtual machines increases in a storage environment, so does the I/O blender effect. Another way to reduce this problem is to put I/O-intensive VMs on higher performance storage.