It’s been over a month since Excelero have added cloud support through Microsoft Azure to their NVMesh software-defined storage solution. Excelero’s NVMesh consists of distributed block storage known for its focus towards I/O intensive workloads such as data analytics, high performance computing and GPU computing for AI / ML / DL.
![](https://i0.wp.com/techunplugged.io/wp-content/uploads/2021/06/image.png?resize=1009%2C374&ssl=1)
According to Excelero, beta users of the Azure-based implementations of NVMesh saw “up to 25x more IO/s and up to 10x more bandwidth to a single compute element, while reducing latencies by 80% from a truly protected storage layer”. It is not clear from the press release about what these performance numbers were being compared to – whether it was an on-premises implementation of NVMesh or a competitor’s storage platform.
Besides performance, one of the benefits claimed by Excelero are better costs on storage due to embedded storage with Azure compute instances which comes at no additional fee. NVMesh should be available on Azure GPU-optimized N-series instances and HPC-optimized H-series instances. Excelero also stated that cloud native workloads will be able to leverage NVMesh through their Kubernetes CSI driver and through integrations with Red Hat OpenShift.
NVMesh will first be available on Microsoft Azure, and will be released later on other major public cloud platforms. To find out more about NVMesh, check their datasheet.
TECHunplugged’s Opinion
This is yet another important development for organizations that are relying more and more on public clouds. Here we see at work one of the outcomes of the COVID-19 pandemic: organizations no longer have the time or luxury to wait for long hardware procurement cycles. Even if they adopt a software-defined storage solution it eventually has to run somewhere, and prolonged or disrupted hardware deliveries naturally makes them look for immediately available options.
Public clouds offer a plethora of storage options, but in advanced fields such as HPC and AI/ML high performance, high throughput and low latency are key; users do not want to rearchitect all of their software and workflows just because a generic solution exists at a given major public cloud. They need to run their jobs or models “here and now”, and the cloud is the fastest way to get this done, especially when the datasets are already present in the cloud.
Although currently limited to Azure, Excelero’s move is a sensible one because it further improves the portability of their solution while making it even more ready for Day 2 operations. The initial release on Azure seems more to be the matter of prioritizing a technology partnership with Azure; we believe the solution should be reasonably portable to other major clouds as stated in the announcement.