Memorize the Future
April 2, 2019
We live in a new age of data, ushered in by the advancements in AI, machine learning and big data analytics. The proliferation of sensors and cameras, all connected by the Internet of Things collects more and more data. In an on-demand economy these data cannot be processed fast enough to allow service providers to respond. The volume, variety, and velocity of data and the need to draw insights from them are unprecedented.
The infrastructure of data struggles to keep up. Modern data-centric applications work with a wide spectrum of data, ranging from batch processing of large files, to stream processing of data on the fly, to ad hoc queries for immediate insights. There has not been a storage system that can perform well with all of these access patterns. To remedy this, many newer data-centric platforms and frameworks are memory-centric, utilizing DRAM as the fastest tier of storage. DRAM has excellent performance characteristics, but it is volatile, and its size is limited by the physics of the media. As a consequence, two of the biggest bottlenecks in an enterprise data center are the performance of storage I/O and the size of memory.
Today marks the beginning of the end to these two pain points. Intel officially shipped Optane DC Persistent Memory, the first of its kind commercially available to the public. This new media consists of novel physical devices that combine the best of both worlds between DRAM and NAND Flash. It is close in speed to DRAM to be used as byte-addressable main memory, and at the same time it is non-volatile like NAND Flash and has 10x size comparing to DRAM.
In the history of computing, memory and storage have been two different concepts. Memory is used to place an application’s running state, accessed by load/store operations, while storage is used to reliably persist data that can survive power cycles. The introduction of persistent memory opens up the possibility of a new infrastructure that unifies memory and storage.
That is the infrastructure we architected. We call this new architecture Memory Converged Infrastructure (MCI). It combines the powers of persistent memory, hyper-converged architecture, container application deployment model and it collapses computing and storage onto a single scale-out cluster of appliances or servers. MCI storage service operates at memory speed, and dramatically accelerates both latency and bandwidth, for both random and sequential accesses. MCI memory service expands the size of memory leveraging the higher capacity of persistent memory as well as memories from other nodes in a cluster interconnected by RDMA. MCI is an architecture that can offer storage services and memory services in tandem, while supporting the backward compatibility of the applications.
Following the MCI architecture, engineers at MemVerge developed a distributed system software called Distributed Memory Objects (DMO) from the ground up. DMO enjoys the advantage of the underlying persistent memory media, and accelerates data-centric workloads dramatically comparing to existing DRAM+SSD/HDD infrastructure. Customers can deploy DMO either as appliances, or as software onto their own servers. Data-centric applications such as Spark, Presto and Tensorflow can be deployed easily on top of DMO, leveraging Kubernetes. We believe DMO will become the infrastructure of choice for many of these applications.
Over the last year, we have been working closely with a number of forward-looking customers, defining our product, and deploying the Alpha version of our product as proofs-of-concept. Both machine learning and data warehousing applications have been deployed on top of our system and we witnessed promising results. Now we are ready to announce the Beta availability of DMO and would like to invite more customers to try out this exciting future!
Talking about the future, we believe the future of storage is memory. We at MemVerge are building a software bridge between that future and the applications of today and tomorrow. This promises to be an exciting journey, and we are proud to announce today that we are taking the first step.
We invite you to join us on this journey!