dcurl - Multi-threaded Curl implementation

Hardware-accelerated implementation for IOTA PearlDiver, which utilizes multi-threaded SIMD, FPGA and GPU.

Introduction

dcurl exploits SIMD instructions on CPU and OpenCL on GPU. Both CPU and GPU accelerations can be enabled in multi-threaded execuction fashion, resulting in much faster proof-of-work (PoW) for IOTA Reference Implementation (IRI). Additionally, dcurl also supports the FPGA-accelerated solution further described in docs/FPGA-ACCEL.md

Warning

• You need to configure paths and flags of OpenCL installation in mk/opencl.mk
• dcurl will automatically configure all the GPU divices on your platform.
• Check JDK installation and set JAVA_HOME if you wish to specify.
• If your platform doesn't support Intel SSE, dcurl would be compiled with naive implementation.
• For the IOTA hardware accelerator, we integrate Lampa Lab's Cyclone V FPGA PoW into dcurl. Lampa Lab provides soc_system.rbf only for DE10-nano board. You need to synthesize to get soc_system.rbf for using Arrow SoCKit board and this RBF file can be downloaded from our release. Moreover, you need to download Lampa Lab-provided Linux image to flash into the micro-SD card and root password is 123456. Finally, you also need to download dcurl into root directory.

Build Instructions

• dcurl allows various combinations of build configurations to fit final use scenarios.
• You can execute make config and then edit file build/local.mk for custom build options.
  • BUILD_AVX: build Intel AVX-accelerated Curl backend.
  • BUILD_GPU: build OpenCL-based GPU accelerations.
  • BUILD_JNI: build a shared library for IRI. The build system would generate JNI header file downloading from latest JAVA source.
  • BUILD_COMPAT: build extra cCurl compatible interface.
  • BUILD_FPGA_ACCEL: build the interface interacting with the Cyclone V FPGA based accelerator. Verified on DE10-nano board and Arrow SoCKit board.
  • BUILD_STAT: show the statistics of the PoW information.
• Alternatively, you can specify conditional build as following:

$ make BUILD_GPU=0 BUILD_JNI=1 BUILD_AVX=1

Test

• Test with GPU

$ make BUILD_GPU=1 check

• Expected Results

*** Validating build/test-trinary ***
        [ Verified ]
*** Validating build/test-curl ***
        [ Verified ]
*** Validating build/test-durl ***
[dcurl] Implementation GPU (OpenCL) is initialized successfully
        [ Verified ]
*** Validating build/test-multi_pow ***
        [ Verified ]
*** Validating build/test-pow ***
GPU - OpenCL
[dcurl] Implementation GPU (OpenCL) is initialized successfully
Success.
        [ Verified ]

• Test with AVX but no GPU

$ make BUILD_AVX=1 check

• Expected Results

*** Validating build/test-trinary ***
        [ Verified ]
*** Validating build/test-curl ***
        [ Verified ]
*** Validating build/test-dcurl ***
[dcurl] Implementation CPU (Intel AVX) is initialized successfully
        [ Verified ]
*** Validating build/test-multi_pow_cpu ***
        [ Verified ]
*** Validating build/test-pow ***
CPU - AVX
[dcurl] Implementation CPU (Intel AVX) is initialized successfully
Success.
        [ Verified ]

• Test with AVX and show the PoW statistics

$ make BUILD_AVX=1 BUILD_STAT=1 check

• Expected Results

*** Validating build/test-trinary ***
        [ Verified ]
*** Validating build/test-curl ***
        [ Verified ]
*** Validating build/test-dcurl ***
[dcurl] Implementation CPU (Intel AVX) is initialized successfully
        [ Verified ]
*** Validating build/test-multi-pow ***
[dcurl] Implementation CPU (Intel AVX) is initialized successfully
        [ Verified ]
*** Validating build/test-pow ***
CPU - AVX
[dcurl] Implementation CPU (Intel AVX) is initialized successfully
Hash count: 3182602
PoW execution time: 0.434 sec
Hash rate: 7333.736 kH/sec
Success.
        [ Verified ]

• Test with Arrow SoCKit board

root@lampa:~# sh init_curl_pow.sh 
root@lampa:~# cd dcurl
root@lampa:~/dcurl# make BUILD_STAT=1 BUILD_FPGA_ACCEL=1 check

• Expected Results

*** Validating build/test-trinary ***
        [ Verified ]
*** Validating build/test-curl ***
        [ Verified ]
*** Validating build/test-dcurl ***
[dcurl] Implementation CPU (Pure C) is initialized successfully
[dcurl] Implementation FPGA is initialized successfully
        [ Verified ]
*** Validating build/test-multi_pow ***
        [ Verified ]
*** Validating build/test-pow ***
CPU - pure C
[dcurl] Implementation CPU (Pure C) is initialized successfully
Hash count: 836032
PoW execution time: 43.000 sec
Hash rate: 19.443 kH/sec
Success.
FPGA
[dcurl] Implementation FPGA is initialized successfully
Hash count: 5125680
PoW execution time: 0.152 sec
Hash rate: 33734.938 kH/sec
Success.
        [ Verified ] 

Tweaks

• Number of threads to find nonce in CPU
  • $ export DCURL_NUM_CPU=26

Performance

After integrating dcurl into IRI, performance of <attachToTangle> is measured as following.

• Each sampling is measured with 30 transaction trytes and total 200 samples are measured.
• mwm = 14, 26 CPU threads to find nonce
• Settings: enable 2 pow tasks in CPU, 1 pow tasks in GPU at the same time

IRI Adaptation

Modified IRI accepting external PoW Library Supported IRI version: 1.5.3

• $ cd ~/iri && mvn compile && mvn package
• $ cp ~/dcurl/build/libdcurl.so ~/iri
• $ cd ~/iri && java -Djava.library.path=./ -jar target/iri.jar -p <port> --pearldiver-exlib dcurl

IOTA PoW Node

gagathos/iota-gpu-pow

• You can construct a IOTA PoW node, which uses ccurl by default
• Generate a drop-in replacement for ccurl and acquire performance boost!
  • $ make BUILD_COMPAT=1 check
  • $ cp ./build/libdcurl.so <iota-gpu-pow>/libccurl.so

Licensing

dcurl is freely redistributable under the MIT License. Use of this source code is governed by a MIT-style license that can be found in the LICENSE file.