This experiment corresponds to Section 6.1, paragraph 1, Figures 11 and 13a of the paper. The goal is to demonstrate Shortstack's linear throughput scalability.
Note for evaluators: In the paper, we use two workloads for this experiment --- YCSB-A and YCSB-C. In the interest of minimizing evaluator effort, the following instructions are to reproduce the results for the YCSB-A workload (this is the worst-case workload among the two for Shortstack as seen in Figure 11c of the paper). The scalability trends for the YCSB-C workload are nearly identical. Nevertheless, if an evaluator would like to also run YCSB-C, a description of how to do so can be found towards the end of this README.
(These are the common preparation steps across most experiments (exp1-4). For a more detailed description of what each of these steps does, refer to this. If you have already undertaken these steps recently while running a previous experiment, then you can skip to directly running the experiments on the client VM)
SSH into Gateway VM (see here for details)
ssh -i shortstack.pem ubuntu@gateway_vm_ip
Switch to the shortstack-artifact folder (this repo), and pull the latest changes
cd shortstack-artifact
git pull origin master
Check the status of the available VMs in the cluster.
cd scripts/aws
./check-instances.sh
For this experiment we need a total of 33 VMs. Hence, make sure that the number returned by the above is greater than or equal to 33 before proceeding. If not, please contact us.
Generate the hosts file, and obtain the IP address (public, private) of the client VM. Make note of these IP addresses.
./gen-hosts.sh ~/shortstack.hosts
Copy the hosts file to the client VM. Replace <client-vm-private-ip> with the private IP address obatined from the previous command.
scp ~/shortstack.hosts ubuntu@<client-vm-private-ip>:/local/deploy/shortstack.hosts
Exit gateway VM (can be done using exit command). This is important. Please do not run actual scripts for experiments on the Gateway VM. Beyond this point, all following steps/commands are to be performed on the client VM.
SSH into client VM (you should have obtained it's public IP address in the previous step)
ssh -i shortstack.pem ubuntu@<client-vm-public-ip>
Switch to shortstack-artifact and pull the latest updates
cd shortstack-artifact
git pull origin master
Run the following to push binaries to all the VMs. This will ensure that all the VMs have up-to-date copies of all application binaries.
./scripts/push_bins.sh
To ease evaluator burden, we have prepared scripts that automate the entire process of running the experiements. For each data point, the following script will take care of both setting up and initializing the cluster (key-value store, proxy servers) and running the clients which will execute the workload. Keep in mind that some steps of the inialization process can take time (upto a minute), and as a result you may see the script pause at certain points of time.
cd exp1
./run_exp1.sh exp1-ycsb-a /local/workloads/ycsb-a-1m
Once the above is complete, you can use this to collect and print the results for all the data points in readable form.
./collect_exp1.sh exp1-ycsb-a
If you would like to run the experiments with the YCSB-C workload, then you can simply repeat the above steps ("Running the experiments" part) with occurences of "ycsb-a" replaced by "ycsb-c".