Sundial

Sundial is a distributed OLTP database management system (DBMS). It supports a number of traditional/modern distributed concurrency control protocols, including WAIT_DIE, NO_WAIT, F1 (Google), MaaT, and TicToc. Sundial is implemented on top of DBx1000.

The following two papers describe Sundial and DBx1000, respectively:

Sundial Paper
Xiangyao Yu, Yu Xia, Andrew Pavlo, Daniel Sanchez, Larry Rudolph, Srinivas Devadas
Sundial: Harmonizing Concurrency Control and Caching in a Distributed OLTP Database Management System
VLDB 2018

DBx1000 Paper
Xiangyao Yu, George Bezerra, Andrew Pavlo, Srinivas Devadas, Michael Stonebraker
Staring into the Abyss: An Evaluation of Concurrency Control with One Thousand Cores
VLDB 2014

Setup Compute Node

Step 0: Edit IP Address and Local Path

• Edit src/ifconfig.txt
  • before "=l", one line corresponds to one compute instance.
  • after "=l", one line corresponds to one storage instance
• Edit info.txt
  • first line: user name
  • second line: absolute path to the repository

Step 1: Setup SSH Key Authorization Locally

• assume there's a public key at /.ssh/id_ed25519.pub
  • generate ssh key link
• make sure each node can access the other node through sudo and you have already tried so that there will not have pop-up question asking about whether to add the ip address
• setup locally: python3 install.py setkey_local

Step 2: Install dependency

• install locally: python3 install.py install_local
• copy current setup to other compute nodes and install remotely python3 install.py install_remote
  • troubleshooting: on each node run "install_local"

Step 3: Setup SSH Key Authorization Remotely

• setup key authorization for each node python3 install.py setkey_remote
  • troubleshooting: command not returned.
    • sometimes first ssh access will spin on requiring input for "The authenticity of host 'node-0 (x.x.x.x)' can't be established. ECDSA key fingerprint is SHA256:xxx. Are you sure you want to continue connecting (yes/no)?"
    • if this is the case, on each node, try executing sudo ssh <other node ip> to bypass the question.

Step 4: More Setups

To set up library path and proto on all compute nodes: python3 install.py config_local python3 install.py config_remote

Setup Storage Service

Edit ifconfig.txt on master compute node, after "=l", use the ip of the (master) storage node (if you use multiple replicas)

Customized Redis Setup

Install redis

git clone https://github.com/redis/redis.git
cd redis
make
cd
mkdir redis_data/

Open redis.conf to setup of Redis:

on Master

• change bind to bind 0.0.0.0 -::1 OR comment out bind to use default
• set up protected-mode (e.g. password), you can use protected-mode no but it is dangerous for public cloud without setting requirepass
• set appendonly yes
• set fsync always
• set dir <your desired path for log>, e.g. dir /users/scarletg/redis_data
• set requirepass sundial-dev

on Replica

• set replicaof <ip of the master> <port>

On Master, start the service

cd src/
./redis-server ../redis.conf

Configuration & Execution

Manual

• Set up config in src/config.h
• For each compute node, do the follow:

./tools/compile.sh
./tools/run.sh -Gn<node_id>

Batch Tests for Experiments

• Different ways of setting up configurations:

  1. directly edit src/config.h
  2. edit .json file and use command line arguments to overwrite some configs.

• If using option 2:

  • template for starting all compute nodes:

  python3 run_exp.py CONFIG=exp_profiles/<name of config file>.json NODE_ID=<current node id> <optional args overwriting config.h>
  

  • example usage:

  python3 run_exp.py CONFIG=exp_profiles/ycsb_zipf.json NODE_ID=0 
  FAILURE_ENABLE=false MODE=debug
  

  • this script allows for list type for each config in .json, and it will enumerate all the possible combinations of the config. i.e. the number of tests in total will be the product of the length of every list.
  • Notes on MODE:
    • if MODE=compile, it will just compile on all nodes without execution
    • if MODE=debug, it will compile and run, but not write output file
    • if MODE=release, it will write output file stats.json to outputs/ after execution

Output

The output file is stored in outputs/. The output data should be mostly self-explanatory. More details can be found in system/stats.cpp.