Home > Storage > PowerScale (Isilon) > Product Documentation > Storage (general) > PowerScale OneFS SmartFlash > OneFS cache architecture
The OneFS operating system caching infrastructure design is based on aggregating the cache present on each node in a cluster into one globally accessible pool of memory. OneFS uses an efficient messaging system, similar to non-uniform memory access (NUMA). This design allows all the nodes’ memory cache to be available to every node in the cluster. Remote memory is accessed over an internal interconnect and has lower latency than accessing hard disk drives.
For remote memory access, OneFS uses either an IP over Ethernet or Sockets Direct Protocol (SDP) over an InfiniBand (IB) back-end interconnect on the cluster; essentially a distributed system bus. SDP provides an efficient, socket-like interface between nodes which, by using a switched star topology, ensures that remote memory addresses are only ever one hop away. While not as fast as local memory, remote memory access is still quick due to the low latency of IB and 40 Gb Ethernet. The relative access latencies of the OneFS caching tiers are covered later in this paper.
Storage nodes can use up to 384 GB of RAM each. A OneFS powered cluster can therefore contain up to 94.5 TB of system memory (252 nodes), plus various SSD configurations for additional read caching.