In its last editorial Processor mention 2 opposite positions about value of defragmentation in virtual infrastructures:
Another issue facing those setting up virtual servers is the increasing disparity between high-performing CPU and memory chips and storage. According to Diskeeper?s Materie, the fastest hard drives available in the mainstream marketplace run at 15,000rpm. In contrast, CPUs and memory are measured in nanoseconds.
According to Materie, server administrators need to look for ways to make the slowest component in the network as fast as possible to alleviate bottlenecks. SMEs can strive for relative parity by employing high-performance hard drive setups. SANs have dropped significantly in price over the last several years, making them affordable for SMEs.
In addition, Materie says that defragmenting storage allows for faster access to data by consolidating file fragments. When the CPU is not searching all over the hard drive?be it virtual or physical?for pieces of a given file, you can expect quicker performance and a decrease in bottlenecks through I/O channels.
For his part, Illuminata?s Haff says that the benefits of defragmenting storage in a virtual setup are minimal, even though virtualization has the potential to cause greater disk fragmentation.
?The problem is that CPU and memory are so much faster than disks that optimizing disk speed is like a drop in the proverbial lake,? Haff says. ?As a last tweak, optimization sometimes makes cost-effective sense, but it is not a problem standing in the way of adopting virtualization.?
Read the whole article at source.
I have to strongly disagree with Haff’s position: in my real-world experience defragmentation of both host and guests disks has a notable impact, being I/O operations a critical bottleneck in every virtual infrastructure.
While it’s true defragmentation isn’t the final solution to performance issues (and I don’t think anybody at Diskeeper or other defrag companies believes so), it’s worth to remember that its price and the effort it requires to work is minimal.
Putting in place a meaningful defrag scheduling (to avoid operations start at workload peeks) brings benefits there is no reason to give up to.