Nimble gets, well, nimble about hyperconverged infrastructure
Is it going to do HCI... who knows?
Interview After thinking that stand-alone external storage array vendors might need to get into the hyperconverged market, we received a tip that Nimble Storage was thinking that too. So we asked Nimble's CEO, Suresh Vasudevan, a few exploratory questions.
Here's the result.
El Reg: How does Nimble view the roles of shared external storage and HCI in offering storage facilities? Do you see HCI growth affecting the external storage market and sales within it?
Suresh Vasudevan: When you look at the external storage market, it is perhaps better viewed as two markets - disk-centric external storage market and Flash-centric external storage market. Of the two, the Flash-centric external storage market is growing very rapidly and we don’t believe is seeing an impact from HCI. The disk-centric storage market is declining because of a number of factors like public cloud, the move to all-flash and alternatives like HCI.
Nimble CEO Suresh Vasudevan
El Reg: What are the strengths and weaknesses of HCI and external storage and how might they complement each other?
Suresh Vasudevan: HCI does have some capabilities that are difficult to match in a best of breed approach including simplicity of installation/setup /provisioning, single-vendor support, and simplicity of procurement. However, there are also some key attributes of external storage that are difficult to match in today’s HCI solutions that include consistency of performance, attractive price-performance, granular scaling, and resiliency in the face of failures / hot-spots.
Given that, we see HCI as well suited for remote offices, smaller-footprint tier 2 environments, and “homogeneous” workloads such as VDI. We see external storage as well suited for tier 1 environments, larger foot-print (capacity or performance footprint) tier 2 environments, and when consolidating multiple heterogeneous workloads.
El Reg: Where do the existing HCI suppliers and offerings fall short?
Suresh Vasudevan: When we look at the current HCI vendors and solutions in the market, we see three short-comings or penalties that need to be overcome for HCI to gain a greater share of the broader opportunity:
- All-Flash penalty - HCI systems have to make two, and often, three copies of data to ensure data durability. While this is less of a penalty in disk-centric HCI systems (disk is cheap), this is a significant cost disadvantage in Flash-centric HCI systems where the media cost relative to the cost of other components is still very significant. Further, current HCI filesystems (which were mostly designed before all-flash became mainstream) have not been designed with good inline data reduction in mind.
Consequently, most HCI systems recommend turning off inline compression and/or deduplication for performance reasons or else they deliver inconsistent application performance if those services are turned on. Lastly, HCI systems mirror all writes to other nodes across the network to guarantee persistence, which can result in poor write latencies.
- Node-sprawl penalty - Whether you want to scale compute, IOPS, throughput or capacity, most HCI systems require that you add more nodes to the cluster. Some of the HCI systems are working to scale some elements independently by using capacity intensive nodes or compute intensive nodes, but they often still require some element of both resource in every node. Over time, this approach results in significant hardware over-provisioning.
- Black box penalty - HCI systems do a good job of simplifying installation, initial setup and provisioning. However, they do not do well at having a detailed and granular understanding of the impact of workloads on underlying system resources, particularly when you have multiple heterogeneous workloads.
For example, if you have a few dozen databases and other workloads on an HCI cluster, and if a database starts experiences high latencies, today’s HCI systems find it very challenging to understand the specific bottleneck that is causing the problem. The answer, as we have heard from numerous customers, tends to be to add more nodes to scale the cluster resources, which may or may not solve the problem.
The black-box penalty also exhibits itself when problems occur. Troubleshooting the root cause can be difficult as there is no visibility into the underlying layers.
It is important to point out that the above penalties are not necessarily a product of HCI as an approach. They result from first generation HCI systems that focused heavily on simplicity of setup and end-to-end provisioning of VMs.
El Reg: If Nimble were to envisage an HCI offering what features and attributes would be important?
Suresh Vasudevan: While we have not announced any product plans around HCI, we believe that we have a very rich foundation:
- The Nimble OS has been designed ground-up to adapt extremely well to flash or even 3DXP over time – for capabilities such as inline compression, inline deduplication, endurance management and so on. It does this in an extremely efficient manner and delivers consistent high-performance.
- The Nimble OS was designed with scale-up, scale-deep and scale-out as inherent attributes of the file-system. Scale-out is necessary to deploy a multi-node HCI cluster, but a more flexible approach to scaling helps address the node sprawl penalty.
- Infosight Predictive Analytics is already very good at monitoring, predicting and preventing issues from VM to storage, modeling resource consumption, predicting resource bottlenecks and recommending corrective actions at an individual VM level. This becomes a key capability in the context of avoiding the black-box penalty.
El Reg:: How might a Nimble HCI offering be developed?
Suresh Vasudevan: Ultimately, the Nimble OS and Infosight are foundational elements of our intellectual property that provide us with multiple avenues to expand our product portfolio over time.
Nimble has not announced hyperconverged product plan but its CEO has kindly provided a list of three areas where current HCI systems can fall short, and then indicated that Nimble IP can help in resolving some or all of these areas.
So is Nimble developing an HCI offering? Your guess is as good as ours, and we think ... it is. ®