Memory Pooling
A key technology in centralized data-center architectures that uses high-speed interconnects such as RDMA and CXL to aggregate the memory of multiple physical servers into a logically unified shared memory pool, allocated dynamically on demand to improve overall utilization.
Timeline (last 90 days)
Memory pooling refers to integrating the memory of multiple physical servers through high-speed interconnects (such as RDMA and CXL) into a logically unified shared memory pool that is dynamically allocated on demand.
The latency of a single remote memory access is typically dozens of times that of local DRAM.
Memory pooling is currently mainly suited to latency-insensitive scenarios such as big data batch processing, in-memory databases, and AI training, and is still hard-pressed to support high-frequency trading or real-time interactive applications.
With CXL, data centers can build memory pooling architectures where multiple compute nodes dynamically allocate memory resources on demand to improve utilization and reduce total cost of ownership.
All Facts (4)
Memory pooling refers to integrating the memory of multiple physical servers through high-speed interconnects (such as RDMA and CXL) into a logically unified shared memory pool that is dynamically allocated on demand.
50%UnverifiedThe latency of a single remote memory access is typically dozens of times that of local DRAM.
50%UnverifiedMemory pooling is currently mainly suited to latency-insensitive scenarios such as big data batch processing, in-memory databases, and AI training, and is still hard-pressed to support high-frequency trading or real-time interactive applications.
50%UnverifiedWith CXL, data centers can build memory pooling architectures where multiple compute nodes dynamically allocate memory resources on demand to improve utilization and reduce total cost of ownership.
50%