We’re receiving more and more inquiries on disk defragmentation for ESX, vSphere and Hyper-V. As IT professionals become more immersed in virtualization, they are beginning to understand how performing vSphere defrag, for example, can lead to vastly improved virtualization performance. It’s not only that fragmentation is a problem in virtual environments because it’s the same ol’ NTFS file system. Fragmentation can be an even greater problem when looking at the relationship of host and guests. IO throughput is the big issue here, and that’s where defragmentation can make a big impact on virtualization performance.
Virtualization promises maximization of server resources, a reduced IT center footprint, a greener solution with respects to energy consumption and reduced hardware investment. All of this can be realized, provided the virtualized servers run as efficiently as possible.
Microsoft’s Technet and other third party sources have documented the negative effects of file and free space fragmentation on physical servers. File fragmentation increases file read access times, while free space fragmentation increases file write times. It is well recognized that fragmentation wastes system resources, increases system maintenance costs, and there is an attendant negative impact on productivity.
The same Windows file system (NTFS) that creates file and free space fragmentation on a physical server is at play inside every virtualized Windows server. The unique issue with virtual servers is that they must share the resources of the physical server on which they reside. To the extent that fragmentation causes one virtual machine to use excess resources, that machine will have a negative impact on the other virtual machines on that physical server.
Earlier this year, Raxco Software met with Scott Drummonds, who at the time was VMware’s top performance expert (Mr. Drummonds has since taken a job with EMC). Scott asked that we perform some testing that would quantify the benefits of fragmentation and he suggested we use VMware’s vscsiStats utility to capture the data. Scott and Raxco agreed on a test format and we performed our tests on an ESX cluster with five (5) virtual machines. The highlights of this testing showed:
• Overall IO Reduction – the PerfectDisk defragmented disk reduced the overall IO count by up to 36.5%. The real overhead with virtualization is the number of IO that must traverse the virtual storage stack. Our testing showed defragmentation was able to reduce this workload by over one-third.
• Overall Disk Write Reduction – the PerfectDisk defragmented disk reduced the total number of disk writes by up to 44.6% when performing an identical task as was done on the fragmented disk. Write performance is a function of contiguous free space availability, and only PerfectDisk consolidates free space within the virtual machine.
• Larger Disk IO – the PerfectDisk defragmented disk produced a great number of the largest IO as measured by the vscsiStats utility (largest measureable size is > 524K). The defragmented disk produced 23 times more IO greater than 524K in size. VMware has a queue depth of 32 for each HBA-LUN pair. If each queue entry represents larger IO, system throughput increases and a potential bottleneck is eliminated.
• Disk Latency – the PerfectDisk defragmented disk improved disk latency by up to 51.3%. Using the vscsiStats data, the defragmented disk substantially reduced the number of IO taking more than 30ms to complete. This is a function of having contiguous files to improve read access time, and contiguous free space to improve write IO speed.
• Application Installation Speed – the testing involved the installation o MS Office and MS SQL. The defragmented disk reduced installation times by 25% and 33% respectively.
Hyper-V and vSphere defrag for improved virtualization performance. Quantifiable results.