Shinjuku is a dataplane operating system project from Stanford. It provides datacenter applications with:
- low-latency (including at the tail)
- preemptive scale that can handle any service time distribution
This work was published on NSDI'19:
Shinjuku is licensed under an MIT-style license. See LICENSE.
Shinjuku requires Intel DPDK and an Intel 82599 NIC. Support for more NICs is on the way.
There is currently no binary distribution of Shinjuku. You will therefore have to compile it from source. Additionally, you will need to fetch and compile the source dependencies:
fetch the dependencies:
./deps/fetch-deps.sh sudo apt-get install libconfig-dev libnuma-dev
Build the dependencies, set up the environment, and run Shinjuku:
cp shinjuku.conf.sample shinjuku.conf # modify at least host_addr, gateway_addr, devices, cpu, and arp address (add client address) ./setup.sh ./build_and_run.sh
Then, try from another another Linux host:
# add arp entry for the Shinjuku server sudo arp -s <IP> <MAC_ADDRESS> ./client/latency_client <IP> <PORT> <RPS> <SPIN_TIME>>