Western Digital, Composable Infrastructure, Hyperscalers, And You

Disclaimer: I recently attended Storage Field Day 19.  My flights, accommodation and other expenses were paid for by Tech Field Day. There is no requirement for me to blog about any of the content presented and I am not compensated in any way for my time at the event.  Some materials presented were discussed under NDA and don’t form part of my blog posts, but could influence future discussions.

Western Digital recently presented at Storage Field Day 19. You can see videos of the presentation here, and download my rough notes from here.

 

Composability and Scale

NVMe-oF

Scott Hamilton (Senior Director of Product Management and Marketing, Data Center Platforms BU) kicked off his presentation by describing composable infrastructure as “AirBnB for storage”. The requirements for data centre storage at scale are increasing exponentially every day. There are a number of challenges when you get to Zettabyte scale:

  • Shared-nothing model strands resources;
  • Lack of agility results in SKU explosion; and
  • New use cases move GPUs to the data.

NVMe over Fabrics (NVMe-oF) provides the solution to these issues. It became a standard in 2016, and provides:

  • Low latency – delivers latencies on par with NVMe SSDs inside x86 servers;
  • High-performance sharing – NVMe-oF attached SSDs can be shared among hundreds of application servers resulting in higher utilisation and lower TCO; and
  • Data access and mobility – Fabric-attached data enables cloud-like dynamic access and workload mobility.

These are the sort of characteristics that enable composable infrastructure to really shine. Hamilton also said that “[b]y 2023, 50% of SSA shipments deployed to support primary storage workloads will be based on end-to-end NVMe technology. Up from less than 2% in 2019”. It seems clear that end-to-end NVMe is what a lot of the kids will be getting into.

Composability and Momentum

Western Digital’s composability infrastructure line has been slowly gaining momentum. OpenFlex is now shipping, and the Open Composable API is available on Open Compute platforms. Western Digital also recently acquired Kazan Networks to accelerate its NVMe ambitions. The Open Composable Interoperability Lab was also recently announced.

 

Thoughts and Further Reading

We tell most folks in the enterprise that they’re not hyperscalers – so they shouldn’t try to behave like a hyperscaler. But a lot of the key infrastructure vendors are heavily focused on servicing the hyperscaler market. It’s a big market, and there’s a lot of money to be had selling to hyperscalers. Why is this important? It strikes me that the concept of composable infrastructure meets a lot of the needs of companies doing compute, storage, and networking at massive scale. Does that mean that Joe Enterprise doesn’t need to worry about how composable infrastructure can help him? Not at all. It just means that some of the early implementations of the technology may not make sense if he’s not operating at a particular scale. The good news is that this architecture will continue to be refined by the likes of Western Digital, and in time we’ll see it adapted to the needs of the enterprise market.

Western Digital presented on a wide range of topics at Storage Field Day 19. There was talk of how the gaming industry was impacting the storage industry, how the Internet of Things was driving development of edge-based infrastructure, and how all of these activities were continuing to push the limits of traditional storage designs. Is composable infrastructure the AirBnB of storage? Maybe, maybe not. Some of this will ultimately depend on the uptake of the architecture in the enterprise and commercial sectors. It’s certainly a super neat concept though, and I think it does a good job of meeting some of the more modern workload needs of infrastructure shops that operate at scale.

I’m looking forward to the day when this kind of technology becomes broadly accessible. The idea of being able to optimise my data centre based on the types of workloads I need to run is extremely appealing. Not every workload is the same, and some things need to run at the edge, some in the cloud, and some in the core. Adopting an architecture in the DC that can adapt to those kind of fluid requirements seems like a great idea. I don’t know that we’re there just yet, but some of that is as much about the maturity of most infrastructure shops as it is about the technology they’re using to serve up workloads to the business. For another view on Western Digital, check out Keiran’s post here, and Chin-Fah posted some interesting thoughts here and here.