This is the second post on the new features included within the Cloud Platform announcement that VMware made last week (first one on licensing can be found here), I will not technically discuss the new stuff, as there are already plenty of posts on the new capabilities, instead I'll try to add my honest comment to them, and my take on how they will enhance (or obstacle) the life of the Infrastructure admin.

Let's start with an updated one: Storage I/O Control (a.k.a. SIOC) is a feature that gives you the ability to automatically throttle the 'noisy neighbors' of your infrastructure on the I/O side and prioritize workloads for IOs in a way that resemble the shares already found in Resource Pools.

SIOC appeared in vSphere 4.1 supporting only block-level datastores and was a very welcomed feature in medium and large-sized environments, especially where there's no physical separation between storage environments and resources are shared, now with vSphere 5.0 all these enhancements are made available also for those using NAS (NFS) datastores.

Then we have Storage DRS (a.k.a. SDRS) that is a brand-new vSphere 5 technology that brings the world-famous VMware Dynamic Resource Sharing to data storage within a virtual environment.

Storage DRS works on two different storage aspects with a common goal, which is to place and keep your VM or VMDK into the right Datastore at the right time, for both performance (guaranteeing low latencies) and availability (guaranteeing enough space to accommodate the VM without running into an out-of-space condition).

Storage DRS uses statistical data for available space and latency metrics coming from the ESX kernel (vmkernel), and if you ever popped the hood of ESX using esxtop you already know that there are lots of storage metrics available in there.

Now, if your job is to deal everyday even with just a moderate quantity of storage estate I can feel your excitement 🙂

But there's a big storage-side caveat in all this.

As I said these new storage features are all fine and good, but they try to understand and predict behaviors that for years storage vendors tried to tweak, enhance and make more intelligent, without resorting to host side techniques, I'm mostly talking about wide striping and automated sub-lun tiering.

For example: if you enable SIOC on a Datastore that resides on a disk pool that is shared between ESX and non-ESX workloads (pretty common in a wide-striping environment) SIOC will start complaining that there is an external workload that is messing with its statistics and calculations, and this behavior can lead to a situation where you will need to choose between using one or the other but not both, not to mention that many people reported that even array-based replication mechanism are known to trigger the error. VMware has a flowchart on how to deal with this kind of issues at their knowledge base that can be reached here: KB 1020651.

On the other side, Storage DRS doesn't play well with the I/O metrics coming from a (proper) sub-lun auto tiered system, as you probably already know, in an ATS storage a single Datastore (or Volume or LUN) can have blocks residing on different tiers of storage, this obviously is completely unpredictable by the host that sees that a specific IOs has been served with a different (sometimes drastically different) latency. This completely screws any prediction and cost-risk-benefit analysis that Storage RDS may do on it, even if there's a new feature called VASA that enhances the dialog between ESX and the storage controller there's no such thing as communicating where the single block is residing when servicing the I/O, the metadata overhead would be too much. Disk space based placement is fine though.

These features, coupled together, are giving VMware admins something that many have just dreamed for in the past, but will these new features force the storage vendors to start providing 'dumb' storage bricks? One move that immediately came to my mind was the Engenio acquisiton made by NetApp, the new line is being touted as a 'Big-data only', 'Hadoop-friendly' and 'for the Media and Broadcast market only' but my opinion is that they will become more and more popular in the future if this pattern of having storage intelligence and QoS on the host side will persist with the new vSphere iterations.