Merge pull request #25 from crunchycookie/multi-node-deployment

add integration and monitoring info

integration-guide.md

@@ -56,7 +56,7 @@ compute-hosts:
    - Make sure `virsh-json` is detected and identified as a command.
 10. Run `./gc-emulator-service conf.yaml` to start the service (consider creating a screen `screen -S gc-emulation-scr` -> run and detach, if you wish to run as a background process - so chances for os to terminate will be limited).
 9. In each compute host, copy `gc-controller` binary from the latest release. Refer core-power-mgt repo and create a config file. Make sure
-core ids and sleep states are correct. Then run `./gc-controller conf.yaml` to start the service, in each compute host. Do the same in the controller node, as we use it as a compute node as well  (consider creating a screen `screen -S gc-emulation-scr` -> run and detach, if you wish to run as a background process - so chances for os to terminate will be limited).
+core ids and sleep states are correct. In the same repo, copy all `.sh` scripts as well. Also, install `virsh-json` via go and copy that binary to the same folder. Then run `./gc-controller conf.yaml` to start the service, in each compute host. Do the same in the controller node, as we use it as a compute node as well  (consider creating a screen `screen -S gc-emulation-scr` -> run and detach, if you wish to run as a background process - so chances for os to terminate will be limited).
 10. Allow controller to discover all nodes with `nova-manage cell_v2 discover_hosts --verbose`
 11. Now all supplementary services are up and running. Lets deploy Openstack. In the controller node, run `./stack.sh`.
 11. Once Openstack is up and running, deploy devstack in each compute node. But for them we use below `local.conf` file.

performance-evaluation/dump-green-score.py

@@ -0,0 +1,20 @@
+import time
+
+import requests as rq
+import sys
+import csv
+
+# 1: local node ip
+# 2: dump csv file.
+print('ip:', sys.argv[1])
+print('dump file:', sys.argv[2])
+print('begin dumping green-score...')
+while True:
+    response = rq.get(url='http://' + sys.argv[1] + ':3000/gc-controller/dev/green-score').json()
+    print('received response:', response)
+    # fields = ['timestamp', 'gs']
+    with open(sys.argv[2], 'a') as f:
+        writer = csv.writer(f)
+        writer.writerow([time.time(), response['green-score']])
+
+    time.sleep(5)

performance-evaluation/monitoring-guide.md

@@ -0,0 +1,40 @@
+#### Monitoring guide
+
+Upon successful integration, the system is ready to be monitored for its performance.
+
+We measure followings.
+
+System metrics,
+
+- Green score
+- Power consumption
+
+Application metrics,
+
+- `nLT`: Normalized lifetime
+- Eviction counts
+
+##### Setting up
+
+In each computing node,
+
+1. Green Score
+   1. `core-power-mgt` service expose an endpoint to obtain green score at a given time: `/gc-controller/dev/green-score`
+   2. In a new linux screen session,
+      1. Run `python3 [dump-green-score.py](dump-green-score.py) <node-ip> <path-to-a-csv-file>`
+      2. csv file should be created and its column names are `timestamp` and `g-score`
+      3. The above script poll endpoint, dumps green score as a new row in the csv file.
+      4. detach from the screen session
+7. Power consumption
+   1. In a new linux screen session,
+      1. Run `sudo turbostat --quiet -out reports/power-stats.csv`. This will keep dumping power stats to the csv file.
+      2. detach from the session.
+
+Application metrics are tracked and logged by the vm arrival trace generation.
+
+##### Post process
+
+In each node, collect power stats and green score csv files. These data needs to be processed and system behaviour 
+can be analyzed.
+
+Once collected, kill all monitoring screen sessions.