SPDK Implementation on a Manycore / Many Node System

Library Content Type:
Publish Date: 
Wednesday, September 29, 2021
Event Name: 
Event Track:

As you know, the Storage Performance Development Kit (SPDK) provides a set of tools and libraries for writing high performance, scalable, user-mode storage applications. Kalray’s MPPA® manycore architecture proposes a unique 80-cores system.

A manycore processor is characterized by an apparent grouping from a software point of view of cores and their portion of the memory hierarchy into computing units. This grouping can delimit the scope of cache consistency and inter-core synchronization operations, include explicitly addressed local working memories (as opposed to caches), or even specific data movement engines and other accelerators. Computing units interact and access external memories and processor I/O through a communi¬cation device that can take the form of a network-on-chip (NoC).

The advantage of the manycore architecture is that a processor can scale to massive parallelism by replicating the computing units and extending the network on chip, whereas for a multi-core processor the replication applies to the core level. For storage purposes, the internal processor clusters are configured with one dedicated cluster as a control and management plane, and the remaining four clusters as four independent data planes.

We have implemented SPDK so that it provides a unique scalable platform that can deliver high performances on an 80-core system.This presentation will explain how we have ported SPDK on our processor core, and what unique pieces of technologies have been developed in order to coordinate with the processor internals. We will also explain how the platform can scale.

  • A platform to deliver unmatched performances with SPDK
  • SPDK performs on a manycore processor up to 80-cores
  • A unique scalable SPDK implementation leveraging manycore efficiency
  • A programmable and opened storage platform on a DPU

Watch video: