P2P bandwidth Example

This is simple example to test data transfer between SSD and FPGA.

KEY CONCEPTS: P2P, SmartSSD, XDMA

KEYWORDS: XCL_MEM_EXT_P2P_BUFFER, pread, pwrite

PCIe peer-to-peer communication (P2P) is a PCIe feature which enables two PCIe devices to directly transfer data between each other without using host RAM as a temporary storage. The latest version of Vitis PCIe platforms support P2P feature via PCIe Resizeable BAR Capability

In this Example we highlight the data transfer between SSD and FPGA and capture the performance numbers for p2p transfer where the input path(/dev/nvme0n1) is provided using -p argument.

Following is the real log reported while running the design on U2 platform:

root@hem-PowerEdge-T640:~# ./p2p_bandwidth -x ./bandwidth.xclbin -p /dev/nvme0n1
Found Platform
Platform Name: Xilinx
INFO: Reading ./bandwidth.xclbin
Loading: './bandwidth.xclbin'
Trying to program device[0]: xilinx_u2_gen3x4_xdma_gc_base_1
Device[0]: program successful!
############################################################
                  Writing data to SSD
############################################################
INFO: Successfully opened NVME SSD /dev/nvme0n1

Map P2P device buffers to host access pointers

Start P2P Write of various buffer sizes from device buffers to SSD

Buffer = 4.00 KB Iterations = 262144 Throughput = 0.16GB/s
Buffer = 8.00 KB Iterations = 131072 Throughput = 0.25GB/s
Buffer = 16.00 KB Iterations = 65536 Throughput = 0.37GB/s
Buffer = 32.00 KB Iterations = 32768 Throughput = 0.50GB/s
Buffer = 64.00 KB Iterations = 16384 Throughput = 0.68GB/s
Buffer = 128.00 KB Iterations = 8192 Throughput = 0.82GB/s
Buffer = 256.00 KB Iterations = 4096 Throughput = 0.92GB/s
Buffer = 512.00 KB Iterations = 2048 Throughput = 0.96GB/s
Buffer = 1.00 MB Iterations = 1024 Throughput = 0.99GB/s
Buffer = 2.00 MB Iterations = 512 Throughput = 1.00GB/s
Buffer = 4.00 MB Iterations = 256 Throughput = 1.02GB/s
Buffer = 8.00 MB Iterations = 128 Throughput = 1.03GB/s
Buffer = 16.00 MB Iterations = 64 Throughput = 1.03GB/s
Buffer = 32.00 MB Iterations = 32 Throughput = 1.04GB/s
Buffer = 64.00 MB Iterations = 16 Throughput = 1.04GB/s
Buffer = 128.00 MB Iterations = 8 Throughput = 1.04GB/s
############################################################
                  Reading data from SSD
############################################################
INFO: Successfully opened NVME SSD /dev/nvme0n1

Map P2P device buffers to host access pointers

Start P2P Read of various buffer sizes from SSD to device buffers

Buffer = 4.00 KB Iterations = 262144 Throughput = 0.05GB/s
Buffer = 8.00 KB Iterations = 131072 Throughput = 0.17GB/s
Buffer = 16.00 KB Iterations = 65536 Throughput = 0.31GB/s
Buffer = 32.00 KB Iterations = 32768 Throughput = 0.54GB/s
Buffer = 64.00 KB Iterations = 16384 Throughput = 0.87GB/s
Buffer = 128.00 KB Iterations = 8192 Throughput = 1.21GB/s
Buffer = 256.00 KB Iterations = 4096 Throughput = 1.55GB/s
Buffer = 512.00 KB Iterations = 2048 Throughput = 1.37GB/s
Buffer = 1.00 MB Iterations = 1024 Throughput = 1.59GB/s
Buffer = 2.00 MB Iterations = 512 Throughput = 1.99GB/s
Buffer = 4.00 MB Iterations = 256 Throughput = 2.15GB/s
Buffer = 8.00 MB Iterations = 128 Throughput = 2.26GB/s
Buffer = 16.00 MB Iterations = 64 Throughput = 2.28GB/s
Buffer = 32.00 MB Iterations = 32 Throughput = 2.33GB/s
Buffer = 64.00 MB Iterations = 16 Throughput = 2.35GB/s
Buffer = 128.00 MB Iterations = 8 Throughput = 2.36GB/s
TEST PASSED

EXCLUDED PLATFORMS:

  • All Embedded Zynq Platforms, i.e zc702, zcu102 etc

  • All Versal Platforms, i.e vck190 etc

  • AWS VU9P F1

  • All Platforms with 201910 Version

  • Alveo U25 SmartNIC

  • Alveo U30

  • Alveo U50 gen3x16 xdma 2019 Version

DESIGN FILES

Application code is located in the src directory. Accelerator binary files will be compiled to the xclbin directory. The xclbin directory is required by the Makefile and its contents will be filled during compilation. A listing of all the files in this example is shown below

data/sample.txt
src/bandwidth.cpp
src/host.cpp

Access these files in the github repo by clicking here.

COMMAND LINE ARGUMENTS

Once the environment has been configured, the application can be executed by

./p2p_bandwidth -x <bandwidth XCLBIN> -f ./data/sample.txt