Rocksteady.md

Run Migration

Start Source Redis-server

In source (netx7):

ps aux | grep redis # check existing redis-server
sudo kill -9 xxxx (redis pid)
cd ~; redis-server --protected-mode no --port 6380 --save "" --appendonly no &

Start Destination Redis-server

In destination (netx5):

ps aux | grep redis # check existing redis-server
sudo kill -9 xxxx (redis pid)
redis-server --protected-mode no --port 7380 --save "" --appendonly no&

Start source priority pull

In source server (netx7), start source migration agent's gRPC server:

cd $NetMigrate/cpp/server/Rocksteady
bash run_src_pull.sh 

Start destination migration agent

In destination server (netx5), start destination migration agent:

cd $NetMigrate/cpp/server/Rocksteady
bash run_dst.sh 

Run YCSB Clients

In client server (netx4):

cd $NetMigrate/cpp/YCSB-client
./ycsb-rocksteady -run -db KV -P workloads/workloadb -P Rocksteady/run.properties -p threadcount=8 -s > ~/result/rocksteady-b-100.txt

Start source migration push to migrate data

After running client for a while (e.g., ~200 seconds), in another terminal in source server (netx7), start source migration agent:

cd $NetMigrate/cpp/server/Rocksteady
bash run_src_push.sh 

After migration finishes, you will get the raw data output ~/result/rocksteady-b-100.txt in client server. You can draw throughput and latency figures from it using $NetMigrate/experiment_steps/figures/draw.py. The trend in the figures will be similar as the below examples.

cp ~/result/rocksteady-b-100.txt $NetMigrate/experiment_steps/figures/
cd $NetMigrate/experiment_steps/figures/
mkdir latency_fig
mkdir thorughput_fig
python3 draw.py rocksteady b 100

Throughput:

Median latency:

99%-tail latency:

Limit Source Redis CPU

If limit source Redis CPU to mimic load-balancing scenario, e.g., 70% and 40% source redis CPU limit:

use this:

ps aux | grep redis
cpulimit -p 1234 -l 70

ps aux | grep redis
cpulimit -p 1234 -l 40