It seems like it should be easy to monitor a wide variety of metrics on the storage array, but in practice it is typically difficult, complex and expensive. Add-in the relationship between virtual workloads and the raw information coming from storage, and things really become sticky. It is this correlation piece, the understanding of the entire virtualization stack, that truly shines on any array that is VM-aware.
Don’t Fight Fires, Win Wars
Many issues in the virtual data center require cross-examination. If a workload or application suddenly experiences high latency, for example, the investigation could branch out in a number of directions. For example:
The array can (and should) be able to either solve or strongly suggest root causes for these issues. As a virtualization administrator, I can attest to the amount of black holes I’ve worked through to understand a seemingly storage-related problem, only to discover that something like spanning tree was giving the packets a good workout and sending them across the WAN and back to my host.
In the world of VM-aware storage, a situation would be revealed by a workload showing latency issues on the network, while disk access looks fine. After all, if it is aware of the virtualization layer, it should also be able to know the time difference between when it sends the data, and when the virtual workload received it.
The Storage Array Knows Best
Where better to go than to the storage array for information on (you guessed it) — the storage array? My experience with array monitoring has been a mixed bag. Technologies like SMI-S and SNMP are common for exporting data, such as IOPS, latency, hit/miss ratios and disk activity out of an array. They typically require a proxy appliance or server to pull the data, and sometimes an additional one to interpret it. This can seem too complex for many use cases, mainly in the name of the elusive single-pane-of-glass view that is so often touted.
Realistically, a complex enterprise environment makes tradeoffs if they want a single management pane to handle a variety of vendor products — you gain a single portal, but often lose vendor-specific, valuable data. No one tool can keep up with all of the improvements that occur with all vendors all at once, and no general specification will meet the needs of all use cases.
Storage Administrators Want In
The day of silos of knowledge that rarely interact are fast coming to an end. And, realistically, a good storage administrator also wants to know what is going on with the virtual machines using the storage. With more unstructured, random workloads hitting the array, it can be quite difficult to track down an issue. A unified team of skill sets able to interact using a common set of tools and information can lead to higher availability and happier users.
VM-aware storage can to assist in many activities beyond firefighting and tuning. A good, clear picture of the end-to-end virtualization stack from the storage perspective also gives a storage administrator a better grasp of the performance, sizing and growth aspects of the arrays. Additionally, many workloads arrive in an unknown state to the storage admin, such as a new application deployed without historical data. Having some hard numbers to trend against can reveal IOPS-per-user metrics, which can be handed to the business to justify IT spending, and also translates things to a simpler number format.
Unique control with VM-level actions for infrastructure functions including snapshots, replication and QoS make protection and performance certain in production, and accelerate test and development cycles.