Where are the storage bottlenecks that can affect virtual machine (VM) performance?
Troubleshooting storage performance problems requires a basic understanding of common storage bottlenecks. For our purposes, there are bottlenecks within the physical disk itself and outside of the physical disk in the interface or network.
Bottlenecks within the disk are related to the mechanical limitations of rotating magnetic storage. The first consideration when troubleshooting storage performance problems is disk speed, which defines how quickly the platters rotate and the latency involved in turning the platters under fixed read/write heads. Faster spinning disks mean lower rotational latency; this is the reason high-performance storage systems use high-speed disks operating at 15,000 rpm. Slower disks (such as 5,000 or 7,000 rpm drives) can introduce excessive storage latency when they are accessed.
Consider the disk's cache size. Since a server's memory and CPU can operate much faster than a mechanical disk, a modest amount of high-speed memory is added to the disk to cache read/write data. For example, data written to the disk is stored in the cache until the mechanical disk can catch up, while data read from the disk is cached until it can be taken by the workload or host system. Larger caches are generally good for disk performance.
Another bottleneck to consider when troubleshooting storage is the disk interface itself. For example, Serial-Attached SCSI (SAS) disks may use an established 3 Gbps interface, but faster 6 Gbps disks and interfaces are available. Moving to a faster interface can ease storage bottlenecks in local disk storage, but it's important to balance faster interface speeds with correspondingly larger cache sizes and rotational speeds. Also, it's usually poor practice to mix faster and slower disks on the same interface because the slower disks will waste the performance potential of the faster interface and faster disks.
When storage is handled in a centralized storage area network (SAN), consider the potential impact of network bottlenecks somewhere between the server and the SAN. For example, a consolidated server running multiple VMs may simply not have enough network ports to support the communication needs of the workloads. Adding network ports or shifting network-intensive workloads to other servers can help. IT professionals may require network analysis tools to help identify and resolve network bottlenecks between workloads and a centralized storage subsystem.
This was first published in October 2013