Making small adjustments to the hard drive settings of virtual servers, along with some other changes, can deliver better performance with minimal effort.
Data centers that are highly virtualized usually select storage area networks (SANs) for various reasons, including remote replication and easier storage management. But some enterprises would rather keep virtual machines (VMs) and data storage separated by having them run on individual servers. Here are a few ways to optimize storage performance in these distributed environments.
Local storage performance on virtualized servers
The challenge with local storage is that an individual server has limited space to accommodate physical disks. This means an administrator can configure a system for high capacity with modest performance, or high performance with modest storage capacity. It is possible to mix features; for example, providing a few high-performance disks and filling remaining bays with high-capacity, lower-performing disks. But you cannot maximize capacity and performance at the same time. Instead, it's best to understand the unique needs of the server's workloads and provide storage with the corresponding characteristics.
If you need better performance, consider the impact of partition alignment on local drives. Partition alignment represents a feature of new hard drives that use 4 KB sectors instead of traditional 512 byte sectors. This lowers the overhead data that accompanies each sector and makes more efficient use of disk space, but the OS must align the drive so sector boundaries to that the disk and OS agree -- otherwise the disk will need to reread and rewrite the larger sectors, which can diminish disk performance. Proper alignment can boost effective disk performance and help some RAID types (such as RAID 5).
Also remember that disk fragmentation can become a problem on local disks, where it is not a problem in a SAN environment. Fragmentation occurs due to the way file systems work. When files are erased and rewritten, the new data can be written into space that was freed as other files were deleted or changed. This behavior is normal, but has a tendency to scatter data around on a drive, and it gets worse over time. The more a file is scattered, the harder a drive has to work to locate all the pieces and the slower it performs. If drive performance appears to degrade over time, it might be worth defragmenting the drives in order to relocate the files so all the pieces are contiguous.
The hypervisor might also be sensitive to local disk storage tasks. For example, features like dynamically expanding disks, differencing disks and snapshots may adversely impact local disk performance -- yet might not have had an effect if SAN storage had been used. If the local server workloads require disk-based protection, it may be necessary to use other local disks (or a SAN) for those tasks. For example, if a VM is stored on disk 1, also taking snapshots to disk 1 might cause notable disruptions to that VM, so try taking snapshots to disk 2 instead, or send the snapshots to an iSCSI SAN.