Scanrail - Fotolia

News Stay informed about the latest enterprise technology news and product updates.

Plexistor SDM Community Edition to converge memory, storage

Plexistor claims its 'software-defined memory' platform lets in-memory databases and traditional enterprise workloads run without dedicated compute-storage clusters.

Venture-backed Israeli startup Plexistor made its debut Tuesday with a software platform that uses nonvolatile memory as primary storage.

Plexistor describes its product as software-defined memory (SDM) that provides high-capacity persistent storage at near-memory speed. The vendor claimed it allows storage for both in-memory databases and traditional enterprise applications to run without dedicated compute-storage clusters.

The inaugural release is called Plexistor SDM Community Edition, which is available as a free download from the company's website. Each installation of SDM Community Edition supports a single node in a shared-nothing configuration that runs on bare-metal servers or Amazon Web Services. 

The company said it has received undisclosed funding from Lightspeed Venture Partners and Battery Ventures. Company CEO Sharon Azulai said Plexistor decouples storage from the data path to reduce data center management and costs.

Plexistor SDM Community Edition presents memory and storage to applications as a file system under a single namespace, although Azulai stressed his company's technology does not virtualize multiple devices as a shared resource pool.

Apps use memory, storage resources as needed

Sharon Azulai, CEO at PlexistorSharon Azulai

Plexistor SDM Community Edition software uses a tier of nonvolatile DIMM memory devices, SATA or NVMe flash drives as a secondary tier for scalable storage, and a tier of spinning disk for less active storage. The Plexistor file system moves data between the memory and storage layers. The single namespace "means you can access your files as if they were local, even though they are going over the network," Azulai said.

The Plexistor SDM translation layer appears as a file system to traditional applications and as byte-addressable storage to in-memory applications. The endgame is to enable individual applications to consume resources as needed.

"We're enabling each application to experience the infrastructure as it needs it. If the application needs memory, it sees the infrastructure as a large memory device. If it needs low-latency storage, it's going to see the infrastructure as a low-latency storage device," he said.

He said Plexistor's approach helps boost the performance of in-memory database applications, such a Cassandra, Kafka, MongoDB and others.

"We are not using memory devices as a cache in any way. We use memory as the primary tier for high performance," Azulai said.

Will Plexistor parlay need for app speed into market share?

The startup's chief challenge is its route to market, which depends on NVDIMM device makers and server vendors integrating its software. The startup has yet to publicly announce vendor partnerships. Azulai said beta customers are running Plexistor SDM Community Edition, but "we don't have customers running it in production."

Arun Taneja, president of IT consulting firm Taneja Group in Hopkinton, Mass., said Plexistor is tapping into growing use of in-memory applications while seeking to reduce data center management and complexity.

The Plexistor SDM translation layer appears as a file system to traditional applications and as byte-addressable storage to in-memory applications.

"Plexistor is trying to come with a single architecture that would use both memory and storage appropriately, according to the needs of the applications. It's a very heady, very novel and extremely complex thing to take on. I visualize this as tiering being stretched from one end of the compute spectrum to the other, starting on the memory side all the way to flash and disk," Taneja said.

Plexistor SDM Community Edition is available as a free download.

Azulai said a paid subscription model will be available later this year, with preliminary pricing of approximately $2,000 per node per year and support planned for storage containers and virtualization.

Next Steps

In-memory processing can be faster, but watch out for bottlenecks

Storage challenges for in-memory databases

Choose the best in-memory database for your needs

Dig Deeper on Solid-state storage