- We use three nodes (node 7, 8, 9) and one switch (sw1)
- Node 7 runs client (Caladan), node 8 runs tcpdump, and node 9 runs server (Capybara)
- (Important) You need to setup some ssh configurations to run this test. Please let me know for this configuration.
- On your $HOME directory
- Run
git clone https://github.com/ihchoi12/caladan.git && cd caladan
- Run
make submodules && make -j
- Run
cd ksched; make ; cd ..
- Run
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- --default-toolchain nightly
- Run
cd apps/synthetic; cargo build --release; cd ../../;
- Run
mkdir /local/$USER/capybara-pcap
- On your $HOME directory
- Run
mkdir Capybara && cd Capybara
- Run
git clone https://github.com/nus-sys/capybara.git && cd capybara
- Run
./scripts/setup/dpdk.sh
- Run
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
- Run
make LibOS=catnip all-examples-rust
- Run
mkdir $HOME/capybara-data
- Open a virtual screen (e.g., tmux).
- Run
cd ~/caladan; sudo scripts/setup_machine.sh; sudo ./iokerneld ias nicpci 0000:31:00.1
on a screen. - Edit
~/Capybara/capybara/eval/test_config.py
according to the test you want to run - Run (once)
pip3 install -r ~/Capybara/capybara/eval/requirements.txt
- Run
cd ~/Capybara/capybara/eval
- Run
python3 run_eval.py
- It will run tests as in the python script. For each test, it will print out a summary of result, and store some log files into the
~/capybara-data/
directory. - log files:
- [test_id].be[x]: console output from the backend server be[x] running on node9
- [test_id].client: console output from the client running on node 7
- [test_id].latency: latency CDF
- [test_id].latency_raw: latency of each request
- If you set
TCPDUMP = True
in the~/Capybara/capybara/eval/test_config.py
file, the script will run only the first test, parse the pcap data files, and then terminate the script. - It will store some log files generated from pcap trace into the
node8:/local/$USER/capybara-pcap/
directory - log files:
- [test_id].pcap: original pcap file
- [test_id].csv: pcap data parsed into a csv format
- [test_id].request_times: timestamps of captured request packets in order
- [test_id].response_times: timestamps of captured response packets in order
- [test_id].pcap_latency: latency of each request in order
- Make sure that the capybara-redis repo is in the same parent directory as this repo.
- Run
make redis-server
to compileredis-server
without migration. OR Runmake redis-server-mig
to compileredis-server
with migration. - Run
make run-redis-server
to runredis-server
. By default, Redis uses theredis.conf
config file. Use a custom config file through environment variableREDIS_CONF
(for replication, for instance).
Demikernel is a library operating system (LibOS) architecture designed for use with kernel-bypass I/O devices. This architecture offers a uniform system call API across kernel-bypass technologies (e.g., RDMA, DPDK) and OS functionality (e.g., a user-level networking stack for DPDK).
To read more about the motivation behind the Demikernel, check out this blog post.
To get details about the system, read our paper in SOSP '21.
To read more about Demikernel check out https://aka.ms/demikernel.
Follow these instructions to build Demikernel on a fresh Ubuntu 20.04 system.
export WORKDIR=$HOME # Change this to whatever you want.
cd $WORKDIR # Switch to working directory.
git clone --recursive https://github.com/demikernel/demikernel.git # Recursive clone.
cd $WORKDIR/demikernel # Switch to repository's source tree.
sudo -H scripts/setup/debian.sh # Install third party libraries.
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh # Get Rust toolchain.
./scripts/setup/dpdk.sh
make
make LIBOS=[catnap|catnip|catpowder|catcollar] # Build using a specific LibOS.
make DRIVER=[mlx4|mlx5] # Build using a specific driver.
make LD_LIBRARY_PATH=/path/to/libs # Override path to shared libraries. Applicable to Catnap and Catcollar.
make PKG_CONFIG_PATH=/path/to/pkgconfig # Override path to config files. Applicable to Catnap and Catcollar.
make install # Copies build artifacts to your $HOME directory.
make install INSTALL_PREFIX=/path/to/location # Copies build artifacts to a specific location.
Follow these instructions to run examples that are shipped in the source tree.
- Copy the template from
scripts/config/default.yaml
to$HOME/config.yaml
. - Open the file in
$HOME/config.yaml
for editing and do the following:- Change
XX.XX.XX.XX
to match the IPv4 address of your server host. - Change
YY.YY.YY.YY
to match the IPv4 address of your client host. - Change
PPPP
to the port number that you will expose in the server host. - Change
ZZ.ZZ.ZZ.ZZ
to match the IPv4 address that in the local host. - Change
ff:ff:ff:ff:ff:ff
to match the MAC address in the local host. - Change
abcde
to match the name of the interface in the local host. - Change the
arp_table
according to your setup. - If using DPDK, change
WW:WW.W
to match the PCIe address of your NIC.
- Change
- Save the file.
sudo -E ./scripts/setup/hugepages.sh
For Catnap and Catcollar, you don't need to run with super-user privileges.
# Server-Side
PEER=server TEST=udp_push_pop sudo -E make LIBOS=catnip test-system
# Client-Side
PEER=client TEST=udp_push_pop sudo -E make LIBOS=catnip test-system
For Catnap and Catcollar, you don't need to run with super-user privileges.
# Server-Side
PEER=server TEST=udp_ping_pong sudo -E make LIBOS=catnip test-system
# Client-Side
PEER=client TEST=udp_ping_pong sudo -E make LIBOS=catnip test-system
- Legacy system call API documentation
doc/syscalls.md
- Instructions for running Demikernel on CloudLab
doc/cloudlab.md
cargo doc --no-deps # Build API Documentation
cargo doc --open # Open API Documentation
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.
See CONTRIBUTING for details regarding how to contribute to this project.
This project is a prototype. As such, we provide no guarantees that it will work and you are assuming any risks with using the code. We welcome comments and feedback. Please send any questions or comments to one of the following maintainers of the project:
By sending feedback, you are consenting that it may be used in the further development of this project.