Turn up the Volumez: Storage orchestrating startup exits stealth

Amir Faintuch, Volumez
Amir Faintuch, Volumez

Israeli startup Volumez is loudly exiting stealth with cloud-based storage orchestrating software now available and the appointment of a highly experienced ex-Global Foundries exec as its new CEO, Amir Faintuch.

Volumez, pronounced volumes, is a cloud-delivered service that provisions block storage for containerized applications in response to a customer saying how much performance, resiliency and security they want. The Volumez engine provisions the data path to local on-premises or raw public cloud storage services, such as Amazon EC2 Instance Store, Google Compute Local SSDs, and Azure Ephemeral OS Disks. Volumez claims its technology makes ultra-high performance data services composable, scalable, and universal across public and private clouds.

Incoming CEO Faintuch said: “Emerging use cases in AI/ML and high performance computing are demanding solutions to close the data management gap on public and private cloud. The market is craving a new approach, similar to what Kubernetes did for compute. I’m excited to join Volumez and lead the company to deliver composable data infrastructure innovation that will remove the barriers and unlock a new level of performance.” 

Amir Faintuch, Volumez
Amir Faintuch

Faintuch was until recently an SVP and GM at Global Foundries and an SVP and GM at Intel in the Platform Engineering Group before that. Volumez says he played a pivotal role in the Global Foundries’ IPO in October of 2021, which raised $2.6 billion.

Volumez Field CTO Brian Carmody told B&F in a briefing: “We’ve taken the conventional model of the storage system and we’ve separated the control plane and the data plane into two completely separate things. The control plane runs in our cloud, and it’s delivered as a SaaS product… The data path runs end to end in the same data centre as where the application is running.”

The Volumez control plane, running in the cloud, issues configuration commands to an on-premises Linux server, NVMe JBOF and its component  drives. The Volumez control plane discovers the JBOFs via the REST servers built into them and connects the Linux server to them using NVMe-over-TCP, which is built into the Linux kernel, or iSCSI.

The data plane exists entirely within a customer’s virtual private cloud (VPC) and is 100 percent open source. Volumez points out a difference between an orchestrator and scale-out storage systems. These, it says, create a proprietary data path often with closed source drivers. 

Users deploy the Volumez CSI driver on Kubernetes clusters or Volumez Connector Service on Instances. In Kubernetes environments, the Volumez CSI driver enables management of persistent volumes through the Kubernetes API and tools such as kubectl. Users need only specify a policy name in the PersistentVolumeClaim, and Volumez does the rest automatically.  

The Volumez Connector is an open source, user space service that issues data path configuration commands and monitors the server data path for drift away from the specified. Users have to deploy a Volumez Connector. The control plane issues commands to the customer’s Volumez Connector Service, a user space service that issues data path configuration commands, and monitors the server data path for drift away from the specified performance, resilience and security declarative statements made by the user.

Examples of drift are a server failure, a network link going down, or an NVMe drive dying. Volumez’s control plane detects and automatically remediates in the background. 

Volumez user declaration choices
Volumez user declaration choices

Depending on requirements, the application data path may include features such as compression, erasure coding and thin provisioning for data reduction. Users do not have to explicitly specify these things, with Volumez saying the inherent complexity of data path construction is abstracted behind the simple concept of declarative policies.


Volumez resiliency declarative choices
Volumez resiliency declarative choices

In the on-premises world, Carmody said: “When we compose the data path, which is ultimately about where that application is going to be storing data end to end… we’re connecting that application server to communicate directly with NVMe drives probably in a JBOF in your data centre. So the data path never leaves the customer’s datacenter.” A JBOF (Just a Bunch of Flash) is an SSD enclosure housing NVMe SSDs but not a storage array controller.

The technology is highly scalable. Carmody said: “There are no controllers. So we’re a controller in this architecture. There’s no noisy neighbor problem. They don’t know about one another. And as a result, it is infinitely scalable from a single node all the way up to a planetary scale, hyper scalar because your control plane in the cloud is not just handling me. It’s handling dozens, hundreds of other customers as well.”

Volumez launched on AWS in the fourth 2022 quarter and its Azure launch is imminent. We understand GCP support is on the roadmap.

Carmody said Volumez “has not written a single line of data path code, no RAID drivers, no anything. All we do is we orchestrate primitives that exist in the Linux kernel. And we believe that this model is the future where off-the-shelf Linux itself is the composable infrastructure data plane of the future for the enterprise in the datacenter. It’s just so ridiculously complex that it can’t be configured by humans at scale. So we built a robot, we built an orchestration service that makes this basically look like magic.

“The DevOps engineer when she’s making a provisioning request: all they do is they specify, for example, give me one terabyte of storage for my production database, give it a million IOPS, 500 microseconds of latency, and make it available across three datacenters. Our robot, the algorithm, takes that request, looks at the inventory of NVMe drives across the customer’s enterprise, and immediately composes a solution that says, OK, we’re going to make it perhaps a RAID 10. And we’re going to put three copies over here, three copies in this datacenter zone, and mirror them, or we’re going to do erasure coding, and keep six copies. The user never has to think about RAID or erasure coding or any of these storage topics. Because it’s declarative.”

How does Volumez get apps running up to 10x faster? Carmody said it ensures that one CPU core handles an app’s compute burden: “Volumez runs the app code and storage code on the same core with data ready in the L2 cache when it’s needed by the app.”

He added: “A single NVMe drive can do 100k+ IOPS with uSec latency. Putting cloud SDS or enterprise storage system controllers inline would kill performance and add huge cost. Raw NVMe is cheap and fast, and our stack provides all the data services (snapshots, replication, etc) natively. The only time our stack connects to another cloud storage or enterprise array is for online data migration when a customer is getting off their previous solution.”

A Volumez-provisioned app on premises can do c900K random read IOPS, with 316 microsecs latency (360 microsecs to application).

Faintuch said Volumez will sell its technology through partners at varying levels of he storage stack. We note that Western Digital and Seagate both supply JBOFs, and Toshiba has its Kumoscale devices. He said: “We are signing some key customers as we speak. Some of them we already signed. We haven’t publicly announced it yet.”

Volumez is focusing on provisioning block storage first, thinking that’s 60 to 70 percent of its opportunity, with file provisioning following later.


Volumez was founded in Israel in November 2020 by original CEO and now CTO Jonathan Amit and COO David Benayoun. Amit was the CTO and co- founder (with Gal Naor) of IBM-acquired Storwize, while Benayoun is an ex-exec in Intel’s Platform Engineering Group. Faintuch will establish Volumez’s US-based operations.

Silk, the rebranded and pivoted Kaminario all-flash array startup, also uses Azure ephemeral disks in its public cloud database acceleration offering.