fi_provider(7) Libfabric Programmer's Manual
Fabric Interface Providers
Conceptually, a fabric provider may be viewed as a local hardware NIC driver, though a provider is not limited by this definition. The first component of libfabric is a general purpose framework that is capable of handling different types of fabric hardware. All fabric hardware devices and their software drivers are required to support this framework. Devices and the drivers that plug into the libfabric framework are referred to as fabric providers, or simply providers.
This distribution of libfabric contains the following providers (although more may be available via run-time plug-ins):
- High-speed InfiniBand networking from Intel. See
fi_psm(7) for more information.
- A general purpose provider that can be used on any network that
supports TCP/UDP sockets. This provider is not intended to provide
performance improvements over regular TCP/UDP sockets, but rather to
allow developers to write, test, and debug application code even on
platforms that do not have high-speed networking.
fi_sockets(7) for more information.
- Ultra low latency Ethernet networking over Cisco userspace VIC
fi_usnic(7) for more information.
- This provider uses the Linux Verbs API for network transport.
Application performance is, obviously expected to be similar to that
of the native Linux Verbs API. Analogous to the Sockets provider,
the Verbs provider is intended to enable developers to write, test,
and debug application code on platforms that only have Linux
fi_verbs(7) for more information.
Libfabric provides a general framework for supporting multiple types of fabric objects and their related interfaces. Fabric providers have a large amount of flexibility in selecting which components they are able and willing to support, based on specific hardware constraints. To assist in the development of applications, libfabric specifies the following requirements that must be met by any fabric provider, if requested by an application.
Note that the instantiation of a specific fabric object is subject to application configuration parameters and need not meet these requirements.
- A fabric provider must support at least one endpoint type.
- All endpoints must support the message queue data transfer interface (fi_ops_msg).
- An endpoint that advertises support for a specific endpoint
capability must support the corresponding data transfer interface.
- FI_ATOMIC - fi_ops_atomic
- FI_RMA - fi_ops_rma
- FI_TAGGED - fi_ops_tagged
- Endpoints must support all transmit and receive operations for any
data transfer interface that they support.
- Exception: If an operation is only usable for an operation that
the provider does not support, and support for that operation is
conveyed using some other mechanism, the operation may return
- FI_ENOSYS. For example, if the provider does not support injected data, it can set the attribute inject_size = 0, and fail all fi_inject operations.
- The framework supplies wrappers around the ‘msg’ operations that can be used. For example, the framework implements the sendv() msg operation by calling sendmsg(). Providers may reference the general operation, and supply on the sendmsg() implementation.
- Exception: If an operation is only usable for an operation that the provider does not support, and support for that operation is conveyed using some other mechanism, the operation may return
- Providers must set all operations to an implementation. Function pointers may not be left NULL or uninitialized. The framework supplies empty functions that return -FI_ENOSYS which can be used for this purpose.
- Endpoints must support the CM interface as follows:
- FI_EP_MSG endpoints must support all CM operations.
- FI_EP_DGRAM endpoints must support CM getname and setname.
- FI_EP_RDM endpoints must support CM getname and setname.
- Providers that support connectionless endpoints must support all AV operations (fi_ops_av).
- Providers that support memory registration, must support all MR operations (fi_ops_mr).
- Providers should support both completion queues and counters.
- If FI_RMA_EVENT is not supported, counter support is limited to local events only.
- Completion queues must support the FI_CQ_FORMAT_CONTEXT and FI_CQ_FORMAT_MSG.
- Providers that support FI_REMOTE_CQ_DATA shall support FI_CQ_FORMAT_DATA.
- Providers that support FI_TAGGED shall support FI_CQ_FORMAT_TAGGED.
- A provider is expected to be forward compatible, and must be able to
be compiled against expanded
fi_xxx_opsstructures that define new functions added after the provider was written. Any unknown functions must be set to NULL.
- Providers shall document in their man page which features they support, and any missing requirements.
Future versions of libfabric will automatically enable a more complete set of features for providers that focus their implementation on a narrow subset of libfabric capabilities.
Logging is performed using the FI_ERR, FI_LOG, and FI_DEBUG macros.
- String representing the provider name.
- Provider context structure.
- Log level associated with log statement.
- Subsystem being logged from.
- Always logged.
- Logged if the intended provider, log level, and subsystem parameters match the user supplied values.
- Logged if configured with the –enable-debug flag.