By Zev Rubenstein, NetApp Data Protection Product Marketing Manager
In Part 1 of this series, I stated that NetApp’s storage has always been “software defined,” and the simple proof is how much intelligence we have built into our OS, Data ONTAP, so that all features work seamlessly together.
In Part 2 of this series, I discussed how we extend “software-defined” with management and workflow automation by integrating with third party tools and storage. I also touched briefly on how clustered Data ONTAP takes our SDS roots to the next level by adding non-disruptive operations (NDO) and seamless scale-out. Today, I will elaborate on how we’ve done that.
A major challenge in any environment with finite resources is how to balance expected needs with actual requirements. In the storage world this typically results in over-provisioning storage when demands don’t meet expectations. We solved that problem years ago when we created aggregates and FlexVol volumes. We were the first storage vendor to abstract the RAID groups with aggregates (each aggregate abstracts multiple RAID groups) and create FlexVol volumes associated with aggregates. That software abstraction enables – among other things – the ability to grow and shrink LUNs and volumes at will. Problem solved. (Diagram courtesy of Dr. Dedupe, who explains this in more detail.)
When we move to a clustered storage environment, a new challenge arises: how to rebalance resources across the cluster. The purported value of clustered, scale out storage is that it creates a massive pool that can be allocated as needed. The ability to load balance both access (users) and storage (the users’ data) across nodes is what turns a set of nodes into a cluster. The trick is in how to do that.
We not only resolved this issue in clustered Data ONTAP, but did it in such a way that it maintains all of the feature interoperability that our customers are used to. How? We decoupled data access from data storage. We created the concept of a “Storage Virtual Machine” (SVM) that abstracts the access to the cluster across all of the nodes. In addition, we created an abstraction layer for the underlying storage resources. The result is that an SVM can have volumes residing in any or all of the nodes and can be accessed by any or all the nodes. In fact, only NetApp can abstract data access and services from pooled hardware resources in both SAN and NAS, a core requirement of SDS architectures.
Our architecture also enables Secure Multi-Tenancy: an SVM is a virtual storage construct that appears to the user as a dedicated storage environment. We also can support a heterogeneous storage layer (including non-NetApp storage and cloud storage). Finally, it enables of Non-Disruptive Operations (NDO): there is no storage maintenance activity that requires downtime any more, from load balancing to hardware and software upgrades. In fact, clustered Data ONTAP has NDO as an underlying design philosophy. More on that in my next blog…
Other vendors have implemented their clustering differently and as a result can’t come close to our capabilities. Of course, we’ve been doing SDS for a lot longer than anyone else. Clearly, design matters.