Volumez introduces a new dimension in block storage management

Volumez, a young Israeli storage management company joined the recent IT Press Tour happening in California a few days ago. It was the opportunity to meet several executives and advisors of the company and we had the opportunity to meet Shmuel Shottan, former CTO of BlueArc, or Iri Trashanski, former SVP of Products at Hitachi Vantara in addition to Amir Faintuch, CEO, Brian Carmody, Field CTO, and one other not public at the time I write this post. Amir Faintuch was named CEO last February and the company raised $20 million for a total of $32 in April.

I already wrote about Volumez when they announced their product, you probably read that article as it was a top news on StorageNewsletter April 19.

But we had now a full story presentation in a very interactive session. Let me dig into it.

First, I mentioned storage management, but the company prefers to say Composable Data Infrastructure or CDI, it appears to be more modern, so let use their term. As all applications use, consume and need data, the main idea is to offer a very standard software service, almost natural and obvious, for any kind of workloads, on-premises or in the cloud, in a Linux driven world.

To understand the solution, we need to explain the various components and how they interact together. First we have a control plane running in the cloud that stores all configurations and declarations for all volumes in a Json format. The engine also collects all the details of all SSDs, on AWS it is about EC2 NVMe instance storage disk, brand, capacity... but above IOPS, bandwidth and latency and remaining values for each as some portions of these drives could be already allocated. This is a fundamental piece as the allocator needs this information to pick right partitions to build volumes. Based on this, the orchestrator, with the configuration file with latency, performance or resiliency criteria, groups partitions and assigns various volumes to the hosted LVMs running within each servers Linux OS and connected to the storage via NVMe/TCP. These servers then use these volumes directly and act as the data plane without any special layer that can potentially reduces I/O performance for instance. This is the data path.

And you see the beauty of this approach, there is no layer installed on production servers, no storage controller in the data path, the global SaaS service centrally manages thousands of volumes that can be fully distributed on the planet. It also illustrates why the control and data plane must be decoupled to offer scalability and availability of the service.

Performance results are impressive and Volumez guarantees these following numbers per volume: 1.5M IOPS, 12GB/s of throughput with 300us of latency with mixed 4k random 70%/30% R/W and 2M IOPS for read-only 4k workloads on AWS without any cache.

We'll have more information on June 19th, so let's wait a bit.