Home > Storage > PowerMax and VMAX > Storage Admin > Reliability, Availability, and Serviceability on PowerMax 2500 and 8500 Arrays > Cache
In PowerMax 2500/8500 systems, the cache sub-system is a services layer responsible for allocation of cache/memory resources to the IO flows and internal “non-data” flows within PowerMax Operating System.
Memory utilization has been increased and cache density improved along with providing optimal memory locality for performance-sensitive workloads in the new arrays.
The cache sub-system is also responsible for ensuring control slot/metadata consistency and recovery across any kinds of software/hardware faults.
Cache is mirrored within the node pair, using dual cast on PCIe. Both PowerMax 2500 and 8500 systems CPU uses Inter-Node bus for fast write mirroring across nodes using DMA (Direct Memory Access).
Cache operations on the dynamic fabric uses RDMA (Remote Direct Memory Access). CPUs on the cache nodes are not needed for cache access over fabric.
Fabric Attach Modules (FAMs) use DMA from remote node cache. FAMs offloads interconnect and data access and can also offload remote drive reads (reads into cache on remote nodes).
Memory for write-miss IO is typically allocated out of Mirrored cache and also services reads on cached writes.
Memory for read miss and prefetch is typically allocated out of non-mirrored cache. This reduces vault infrastructure and improves cache efficiency.
There are numerous code scans that work together to ensure that data in cache is clean, and that the infrastructure that supports it is working correctly.
General cache integrity scrubbing also runs on each EM emulation, which validates cache slot linkage, control slot sanity, and verifies transient state slots.
In PowerMax 2500 and 8500, scans are now multi-threaded for greater efficiency and performance.