What are you looking for ?

ISC: MemVerge Unveils CXL-Based Multi-Server Shared Memory with Project Gismo

To accelerate distributed applications by eliminating the IO wall consisting of network IO and data copies

MemVerge, Inc. announced Project Gismo (Global IO-free Shared Memory Objects) at the International Supercomputing Conference (ISC).

This technology changes the performance of distributed applications by eliminating network IO and data copies. An overview of Gismo was presented at the CXL Forum at ISC on May 23. Demonstrations of Gismo were presented at ISC in the company’s booth.

Project Gismo introduces a CXL-based multi-server shared memory architecture, pushing the boundaries of data access and collaboration in distributed environments. By integrating Compute Express Link (CXL) technology, Project Gismo enables real-time data sharing across multiple servers, eliminating the need for network IO and reducing data transfer delays.

Network and storage IO are performance bottlenecks for distributed, data-intensive applications,” said Charles Fan, CEO and co-founder. “Project Gismo leverages the emerging CXL platform and will effectively remove this bottleneck known as the IO Wall. We envision a memory-centric future where memory is the network and memory is the storage.”

This solution empowers organizations to unlock the full potential of their distributed applications, accelerating data-intensive workloads and reducing latency to unprecedented levels. With Project Gismo, enterprises can achieve performance gains while streamlining their infrastructure and optimizing resource utilization.

Key use cases of Project Gismo are AI/ML applications including large language models, next-gen databases, and financial trading platforms.

Ray is an open-source unified compute framework that makes it to scale AI and Python workloads – from reinforcement learning to DL to tuning, and model serving. Learn more about Ray’s rich set of libraries and integrations.

According to Zhe Zhang, head, open source engineering, Anyscale, Inc.Ray is used to scale large language models such as ChatGPT. MemVerge’s Project Gismo extends Ray’s zero-copy-memory-centric architecture across multiple server instances, significantly improving the performance of shuffle and other data exchanges. We look forward to continuing the collaboration with MemVerge and the Ray community.” 

One of the early adopters of Project Gismo is Timeplus, Inc., a developer of a next-gen real-time streaming database. Their use case focuses on using Gismo to improve the fault-tolerance of their database, achieving a 20x improvement in fail-over speed.

According Ting Wang, CEO, Timeplus, Inc.With Gismo’s revolutionary CXL-based multi-server shared memory architecture, we have experienced a remarkable improvement in the fault-tolerance of our database system. The speed of streaming query fail-over has been accelerated by an impressive 20x, allowing us to provide unparalleled user experience for the continuation of data streaming processing.”

MatrixOrigin, a developer of a next-gen ‘Hybrid Transactional/Analytical Processing Engine’, is another early adopter of Project Gismo. It has found that Gismo simplifies its caching architecture and improves performance in both transaction processing and analytical processing in the cloud.

Project Gismo has simplified our caching architecture to a great extent, eliminating the need for complex data copies and reducing data skew,” said Feng Tian, CTO, MatrixOrigin. “The global shared memory accessible by multiple servers has enabled seamless data sharing and collaboration, enhancing the overall efficiency of our engine. As a result, we have experienced a remarkable boost in performance, enabling us to handle larger workloads and deliver faster query responses for our customers.”

The company’s technology is set to redefine the landscape of distributed computing, empowering businesses to scale their operations and maximize productivity. The launch of Project Gismo at ISC marks a milestone in the firm’s commitment to change the industry and drive the future of memory-centric computing.

Read also :