The original Call For Participation for this TC may be found at https://lists.oasis-open.org/archives/members/202211/msg00001.html
(1)(a) TC Name
Infrastructure Data-Plane Function (IDPF) TC
(1)(b) Statement of Purpose
This project defines a standardized, multi-vendor PCIe Network Data-plane Function Device.
Compliant implementations provide high throughput over Ethernet and Remote Direct Memory Access (RDMA), allow live migration and decoupling of Tenants and Data center operator Infrastructure operations.
Devices can be composed in any manner desired, including support for Single-root I/O Virtualization (SR-IOV), Scalable I/O Virtualization (SIOV), or fully/partially emulated devices
The project defines the Host Interface, Device Behavior, setup, and configuration flows. The result can be presented to the Operating system as a physical Function (PF) or a virtual function (VF) PCIe device, as needed.
The purpose of this project is to define a standardized Network Data-Plane Function PCIe device that satisfies multiple goals:
- High throughput/low-overhead operation
- Remove most of the CPU cost of emulated interfaces
- Decouple Tenant and data-center operations
- Zero downtime for Tenants while Datacenter operator is free to bring down server for maintenance
- Allow live migration of tenants and data-center service nodes
- Standardized interface for scalability and ease of configuration and operation
- Offer Tenants to be offered additional features and accelerations to keep up evolving demands
- Feature negotiation and extensibility to allow scalable provisioning and orchestration
None of the standard interfaces available in the market today provide for all of these needs at once.
This project aims to complement rather than supplant the work of the Virtual IO Device (VirtIO) OASIS TC, providing for accelerated development by focusing exclusively on PCI Express hardware implementations (and their equivalent SW emulated devices) and on the networking functionality.
To facilitate and promote cooperation between the two projects, IDPF TC member companies are encouraged to also be members of the VirtIO TC, and Contributors of material comments will also be encouraged to cross-post their comments to the VirtIO TC comments mailing list
This project is aimed at meeting the needs of
- Operators of large data centers (especially public cloud service providers), and their Tenants
- Vendors of Host-attached networking devices (Network Interface Cards (NICs), Accelerators, etc.)
Having a Multi-vendor Standard device allows both sides to scale their efforts, and make them more efficient:
- Vendors can avoid per-customer customization while still allowing feature extensibility and “secret sauce” differentiation
- Datacenter operators will have a scalable solution that can be orchestrated as a generic good, avoiding the need for per-supplier customization and constraints. i.e. make Host-attached networking devices more uniform in treatment, while still getting maximum performance and minimum coupling to a specific vendor or product version/generation.
- Tenants will be able to get high-throughput feature-rich networking connectivity, over standard transports, while being decoupled from the Datacenter operator concerns and constraints
Scope of the TC
- Interface & behavioral model of an IDPF device as seen by an IDPF driver
- Application Programming Interface (API) & Descriptor formats
- Initialization/reset, configuration & Operational Flows
- Base IDPF specification + Capability Discover/Negotiation mechanism
Out of scope for TC work are
- Testing framework for any specific Device’s Compliance
- Code/Specification for any Offloading service not part of the base IDPF specification (i.e. Per-vendor/Per-case Extensions)
The TC will deliver
- A specification of the Interface and Behavioral model of an IDPF-conformant device
- 1st Draft to be submitted to TC soon after formation
- A reference Software (SW) Device
- A reference IDPF driver
- Intel/Google Device driver soon after TC formation
- (if Needed) Driver for Reference Device : 12m after TC formation
(1)(e) IPR Mode
TC will follow the Non-Assertion mode
Networking SW architects & Implementers
Operating Systems (O/S) device Driver Writers
PCIe Network-cards, Infrastructure Processing Units (IPU’s) and Accelerator vendors
Original Equipment Manufadcturares (OEM’s) & Original System/Software Vendors (OSV’s)
TC’s language will be English