Skip to content

Dash User Guide: vSMP Node

The vSMP node is an aggregation of 16 compute nodes with a total of 128 cores and 768 GB of memory. The software interception engine creates a uniform execution environment for applications. The vSMP Foundation creates the relevant BIOS environment to present the OS (and the SW stack above it) as single coherent system. Cache coherency is maintained between boards using multiple mechanisms, on a per-block basis, based on real-time memory activity access pattern. The board local-memory is leveraged for caching and the IB backplane latency is hidden using software-driven adaptive caching techniques. The main caching algorithm families are:

  • DSRAM (Distributed Shared RAM)
    • 4K cache line — memory migration and replication
    • COMA model — cache as-much-as-you-can
    • High one-time synchronization overhead: involves 3 interconnect hops and 2 additional boards.
    • Enhancements: RMW prediction (pre-fetch)
  • LBC (Large Block Copy)
    • Instruction-based large-block pre-fetch, remote-DMA accelerator
    • Leverage access pattern prediction to drive COMA efficiency
  • NASRAM (Node Attached Shared RAM)
    • Instruction-size access — memory pinning
    • NUMA model — no local caching
    • Low remote per-address access cost: involves 2 network hops (1 additional board)
    • Enhancement: Ansyc. write

vSMP exposes all available I/O resources to the OS in a unified PCI hierarchy. On Dash the following aggregated vSMP filesystems are available:

/ramfs (aggregation of ram)
/scrach (aggregation of SSD drives)

Back to Top


Did You Get
What You
Wanted?
Yes No
Comments