Disclaimer: I recently attended Dell Technologies World 2019. My flights, accommodation and conference pass were paid for by Dell Technologies via the Media, Analysts and Influencers program. 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.
As part of my attendance at Dell Technologies World 2019 I had the opportunity to attend Tech Field Day Extra sessions. You can view the videos from the session here, and download my rough notes from here.
One of the presenters at Tech Field Day extra was Liqid, a company that specialises in composable infrastructure. So what does that mean then? Liqid “enables Composable Infrastructure with a PCIe fabric and software that orchestrates and manages bare-metal servers – storage, GPU, FPGA / TPU, Compute, Networking”. They say they’re not disaggregating DRAM as the industry’s not ready for that yet. Interestingly, Liqid have made sure they can do all of this with bare metal, as “[c]omposability without bare metal, with disaggregation, that’s just hyper-convergence”.
[image courtesy of Liqid]
The whole show is driven through Liqid Command Center, and there’s a switching PCIe fabric as well. You then combine this with various hardware elements, such as:
- JBoF – Flash;
- JBoN – Network;
- JBoG – GPU; and
- Compute nodes.
There are various expansion chassis options (network, storage, and graphics) and you can add in standard x86 servers. You can read about Liqid’s announcement around Dell EMC PowerEdge servers here.
Other Interesting Use Cases
Some of the more interesting use cases discussed by Liqid included “brownfield” deployments where customers don’t want to disaggregate everything. If they just want to disaggregate GPUs, for example, they can add a GPU pool to a Fabric. This can be done with storage as well. Why would you want to do this kind of thing with networking? There are apparently a few service providers that like the composable networking use case. You can also have multiple fabric types with Liquid managing cross composability.
Liqid have customers across a variety of workload types, including:
- AI & Deep Learning
- GPU Scale out
- Enable GPU Peer-2-Peer at scale
- GPU Dynamic Reallocation/Sharing
- Dynamic Cloud
- CSP, ISP, Private Cloud
- Flexibility, Resource Utilisation, TCO
- Bare Metal Cloud Product Offering
- HPC & Clustering
- High Performance Computing
- Lowest Latency Interconnect
- Enables Massive Scale Out
- 5G Edge
- Utilisation & Reduced Foot Print
- High Performance Edge Compute
- Flexibility and Ease of Scale Out
Thoughts and Further Reading
I’ve written enthusiastically about composable infrastructure in the past, and it’s an approach to infrastructure that continues to fascinate me. I love the idea of being able to move pools of resources around the DC based on workload requirements. This isn’t just moving VMs to machines that are bigger as required (although I’ve always thought that was cool). This is moving resources to where they need to be. We have the kind of interconnectivity technology available now that means we don’t need to be beholden to “traditional” x86 server architectures. Of course, the success of this approach is in no small part dependent on the maturity of the organisation. There are some workloads that aren’t going to be a good fit with composable infrastructure. And there are going to be some people that aren’t going to be a good fit either. And that’s fine. I don’t think we’re going to see traditional rack mount servers and centralised storage disappear off into the horizon any time soon. But the possibilities that composable infrastructure present to organisations that have possibly struggled in the past with getting the right resources to the right workload at the right time are really interesting.
There are still a small number of companies that are offering composable infrastructure solutions. I think this is in part because it’s viewed as a niche requirement that only certain workloads can benefit from. But as companies like Liqid are demonstrating, the technology is maturing at a rapid pace and, much like our approach to on-premises infrastructure versus the public cloud, I think it’s time that we take a serious look at how this kind of technology can help businesses worry more about their business and less about the resources needed to drive their infrastructure. My friend Max wrote about Liqid last year, and I think it’s worth reading his take if you’re in any way interested in what Liqid are doing.