Experiment

for each latency config  
    for each data structure[0:crdt, 1:opsets, 2:global lock, 3:rwlock]:  
        set lc, exp variable in app.py  
        clear logs  
        run app  
        base load  
        conflict load  
        save results  
        stop app

clear logs

for f in $(find data); do > $f; done

base load

sh workload/base.sh

running 0% conflict workload

sh workload/conflict0_load0.sh & sh workload/conflict0_load1.sh & sh workload/conflict0_load2.sh 

save results

cp -r data/ lc<lc_config>/data0_<api>/

running 10% conflict workload

sh workload/conflict10_load0.sh & sh workload/conflict10_load1.sh & sh workload/conflict10_load2.sh 

save results

cp -r data/ lc<lc_config>/data10_<api>/

running 20% conflict workload

sh workload/conflict20_load0.sh & sh workload/conflict20_load1.sh & sh workload/conflict20_load2.sh 

save results

cp -r data/ lc<lc_config>/data20_<api>/

Results

The experiment has been run on DELL PowerEdge R410 machine, 64 GB RAM, 24 cores @ 2.40GHz Intel Xeon E5645 processor, Linux kernel version 3.16.0-4-amd64.

The result is available at https://nuage.lip6.fr/s/LFzQBWmXTxb36Ap.