Virtualization storage performance has a direct effect on application performance and user experiences, so admins need to implement comprehensive monitoring to track storage performance.
Virtualization storage performance monitoring provides three distinct advantages:
1. Storage performance monitoring provides real-time warnings of serious problems -- usually summoning an administrator or technician when thresholds are exceeded.
2. It yields important historical data about the virtualization storage that could reveal usage trends and help IT departments predict future resource needs before storage performance is affected.
3. Storage performance monitoring helps reveal problems. Virtualization obfuscates the relationship between storage and applications, and that's fine until trouble occurs. Abstraction can make it difficult (or even impossible) to locate the source of storage performance problems, so storage performance monitoring helps cut through that fog. Proper monitoring can help identify the real location of deteriorating virtualization storage performance.
Experts point out that the actual baseline for virtualization storage performance monitoring points are almost identical to those used for traditional (nonvirtualized) storage. These include metrics that should already be familiar to many storage professionals, including LAN and/or SAN bandwidth, I/O operations per second (IOPS), storage queue length, throughput and response time.
Don't give all storage performance metrics the same weight. Palmer says that the importance of each metric in storage performance monitoring will vary by application and server. For example, response time may be most important for a front-end, customer-facing application, while IOPS may be critical for small-block, batch-processing applications.
Storage performance monitoring tools
Virtualized environments are assisted by a new generation of storage management tools and application programming interfaces that are more "virtual machine-aware" than they were in the past. Many storage vendors are now capable of displaying the properties of virtual machines (VMs) from within the storage management system. Previously, storage performance monitoring tools displayed only logical unit numbers (LUNs) and volumes, and there was little or no visibility into the contents of those storage units and into the VMs within them. This often led to difficulty in calculating the true amount of disk space consumed by a VM and identifying which VM was the source of excessive I/O demands to the storage layer.
In addition, great strides have been made in optimizing the demands made by the hypervisor to the virtualization storage layer, and it's now possible to offload a great deal of the IOPS that were generated by common VM tasks such as disk-to-disk backup, taking a snapshot or cloning a VM to the array. The new generation of arrays whose firmware is compatible with VMware's vStorage APIs for Array Integration (VAAI) initiative is a good example.
Even though service levels will vary between different storage tiers or pools, storage performance monitoring should ideally be applied uniformly. This gives a more comparative view of virtualization storage resources.
But experts are quick to note that there are exceptions to this rule about virtualization storage performance.
"There's always a cost consideration," said Dwayne Sye, chief information officer at Cvent Inc., an event management software company in McLean, Va. "If it's cost-prohibitive to monitor a SATA array at the same level as a Fibre Channel array, economics may play a factor. But certainly, if it's possible ... I think that would be the optimal situation."
Tony Palmer, senior engineer and analyst at Enterprise Strategy Group, takes a slightly different approach, underscoring the role of applications rather than virtualization storage. For example, monitoring the storage used for Exchange Server may take a higher priority than monitoring the storage used for backups, resulting in different storage performance monitoring procedures. "It should all be monitored using the same tools, but you're going to change your tactics appropriate to the importance of the application rather than the importance of the storage," he said.
The future of storage performance monitoring
Virtualization admins can look forward to a time when virtualization storage systems support something known as "auto-tiering." This is the ability for LUNs to be "migrated" from one storage array to another or from one class of storage, say SATA, to another, such as solid-state drive. This "Storage vMotion," for want of a better phrase, is already present in some newer storage arrays, and it can eliminate "hotspots" where IOPS are oversaturating a particular array or subsystem.
Solid-state storage is increasingly used tactically to remove the disk spindle from the performance equation. This has opened the door to applications that have historically been regarded as too dependent on disk I/O to be easily virtualized.
As with thin provisioning, customers should expect this to happen seamlessly at the storage layer but controlled and governed at the virtualization layer by a policy-based system. The requirement for administrative approval should stop VMs from being moved unexpectedly. Sadly, this kind of storage vendor-optimized volume or LUN migration will likely be available only in the top tier of enterprise-class storage.
Someday, dynamic resource management systems will include not just optimizations for memory and CPU-bound workloads, but also the capacity to optimize the network and storage I/O paths.