Prepare
Qualify the device, driver, reset behaviour, lender-side peer-to-peer capability, NTB mapping size, IOMMU policy and platform support for the selected access model.
Reference architecture · Managed resources
Shared PCIe Device Access
Pool compatible PCIe devices and expose them through Device Lending with native drivers or through the SISCI SmartIO extension for application-managed access.
Resource pool
NVMe · GPU · FPGA · NIC
Ownership
Exclusive devices; virtual functions may be assigned separately
Software
Device Lending / SISCI SmartIO
Operations
Assign, release, reset and reassign
Reference topology
Device pool and lending
Host A
Borrower / application
Requests an available resource
Host B
Borrower / application
Uses another assigned resource
Host C
Standby / workflow
Receives a device after release
NTB fabric
Direct or switched PCIe fabric
NTB mappings and SmartIO lifecycle control
resource
NVMe · GPU · FPGA pool
Compatible devices presented through supported models
Control plane
PCIe data plane
Operating model
How the architecture works
The sequence describes system behaviour rather than a product feature list.
Assign
Device Lending presents an eligible remote device to one borrower for native driver binding; SISCI SmartIO exposes device resources through an application-managed API.
Release
The workflow quiesces active users, releases or unbinds the device, restores or resets state where required and returns the resource to the pool or local use.
System definition
Reference topology and architectural boundaries
Ownership, data movement, software responsibility and the limits of the pattern are defined separately.
System-level view
Device pool and lending
Host A
Borrower / application
Requests an available resource
Host B
Borrower / application
Uses another assigned resource
Host C
Standby / workflow
Receives a device after release
NTB fabric
Direct or switched PCIe fabric
NTB mappings and SmartIO lifecycle control
resource
NVMe · GPU · FPGA pool
Compatible devices presented through supported models
Qualification required
Application-aware design
Ownership model
Exclusive by default
Device Lending normally assigns a device to one borrowing host at a time.
Concurrent access
Requires explicit support
SR-IOV or an application-managed SmartIO model is needed for supported simultaneous use.
Driver model
Device and release dependent
Native unmodified driver borrowing is documented for Linux; SISCI SmartIO uses an application-managed access path. Device, kernel, platform and release support must be confirmed.
Lifecycle model
Managed assignment
Hot-add, removal, reset, IOMMU grouping and cleanup must be engineered as one workflow.
Engineering criteria
Design decisions that determine whether the architecture will work
These items must be resolved for the actual hosts, endpoints, operating systems, topology and workload.
Access model
Choose Device Lending for exclusive native-driver ownership, SISCI SmartIO for application-managed device access, or separately assigned SR-IOV virtual functions where supported.
Device compatibility
Confirm device class, driver, firmware, reset capability, lender-side P2P support, NTB window requirements and the published platform/kernel support matrix.
Isolation
Review IOMMU groups, requester IDs, ACS behaviour, security boundary and tenant/workflow separation.
Lifecycle automation
Define assignment, quiesce, unbind, reset, handover, failure recovery and audit logging.
Capacity planning
Size device count, fabric ports, lane width, power, cooling and expected concurrency around the workflow schedule.
Implementation layers
Building blocks used to realize the architecture
The final system combines compatible hardware, software, application logic and validation—not one standalone product.
Borrowing hosts
Independent computers with compatible NTB adapters and supported software.
Managed fabric
A direct or switched NTB fabric configured for address translation, resource routing and isolation.
Device pool
Qualified NVMe, GPU, FPGA, NIC or other supported PCIe resources.
eXpressWare services
SmartIO core services, Device Lending, the SISCI SmartIO extension and release-specific management tools.
Workflow control
Scheduler, policy, ownership state, audit and failure handling.
Qualification suite
Assign/release cycles, reset, driver rebinding, load, isolation and recovery tests.
Selection boundary
Where this architecture fits
Use the pattern when its ownership and data-movement model match the engineering requirement.
Use this architecture
Expensive or specialized PCIe devices sit idle because every host has a dedicated copy.
Use another architecture
Several hosts must transparently and concurrently use a device that has no virtualization or application-managed sharing support.
Typical deployments
Shared NVMe pools, accelerator farms, test resources, FPGA appliances and scheduled engineering workflows.
Project definition
Inputs required before system configuration
Architecture selection starts with the actual platform, traffic, software and recovery requirements.
Device inventory
Exact device, firmware, driver, reset method, SR-IOV capability and IOMMU grouping.
Host inventory
OS, kernel/driver release, slots, adapters and security policy.
Ownership policy
Who may borrow, duration, concurrency, priority, release and audit requirements.
Failure handling
Host crash, device timeout, forced reclaim, reset and restoration to service.
Define the topology before selecting the components.
Primionics can review the root-complex model, endpoint inventory, lane and bandwidth budget, software path, operating-system support and qualification requirements for the complete PCIe system.