18
1 Optimizing the Efficiency of the NCAR- Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde Mentor: Aaron Andersen August 1, 2014

1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

Embed Size (px)

Citation preview

Page 1: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

1

Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility

ASoftware Perspective

Theophile Nsengimana Collaborator: Ademola Olarinde

Mentor: Aaron Andersen

August 1, 2014

Page 2: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

2

Project Goals• Work on Software Tools to Automate:

– Collection of NWSC Building Sensor Data

– Quality Control

– Visualization of Key Building Parameters

• Work with Ademola Olarinde:– Software Tools

– Requirements

– Data Exploration

Page 3: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

3

• Preplanned Method:– Store sensor-based raw data generated by Johnson Controls Inc. into a

data store.– sMAP (Simple Measurement and Actuation Profile) from Berkeley. [1]

• Current Method:– Restructuring raw data into CSV file(s).

Data Collection

Page 4: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

4

sMAP Overview• An open source modular software designed fundamentally to ease

the collection, storage and retrieval of time series data. • Time series source: archival and real-time from sensors.• Time series can be tagged with metadata

Page 5: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

5

sMAP Components

• sMAP sources– Connect to physical sensors to expose the data they generated to

sMAP archiver (repository) via http.– Real-time data.

Page 6: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

6

sMAP components

• sMAP archiver– A high-performance data store – Connects to both relational and time series databases. (Postgres for

metadata storage and Readingdb for time series storage)

• Applications– Make use of data: visualization, computing control optimal

strategies, etc.

Page 7: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

7

Challenges with sMAP

• General– Documentation– Community support

• Relatable to this project– Couldn’t load Archival data whose time was out of the range

[(time_now – 24 hours), time_now]

Page 8: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

8

Quality Control

• Proper Formatting– Hourly (or daily, monthly, yearly) Interval as opposed to 15-minutes

interval– Match time format across all generated csv files– Handle missing data– Eliminate irrelevant data– Merge properly formatted csv files into one csv file for faster access

Page 9: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

9

Quality Controldef day_interval(self): BY = ‘day’ init_dt, init_value = self.init_data.popitem() samedts = self.getSameDatetimes(self.init_data, init_dt, BY) #find datetimes with same day as init_dt value = init_value #the sum of values of ‘samedts’ missingCounter = 0 #missing data zeroCounter = 0 #how many times this sensor has been offline. for ndt in samedts: ch_value = self.init_data.pop(ndt) try:

if float(ch_values) != 0: value += float(ch_values) elif: zeroCounter += 1 except ValueError: missingCounter += 1 #calculate the average if missingCounter == len(samedts): value = ‘M’ elif (zeroCounter == len(samedts)) or (zeroCounter > 0 and (zeroCounter + missingCounter == len(samedts))): value = 0 else: value = value / ( len(samedts) – (missingCounter + zeroCounter)) self.fin_data[self.makedt(init_dt, BY)] = value

Page 10: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

10

Data Analysis & Visualization

• Python packages– Matplotlib – Numpy

• Main focus– Statistical correlation– Basic Control Charts– Plots

Page 11: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

11

Statistical Correlation• How a sensor’s performance is related to other various factors.• E.g., Cooling Towers vs Outside Air and Computer Load

Cooling Towers’ correlations against time, wet-bulb, wet-bulb depression, dry-bulb and computer load.

Page 12: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

12

Control Charts

• Whether a particular sensor’s performance is in a statistical control.

• E.g., evaluate the performance of Condenser Water pumps.

Page 13: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

13

Plots

• Visualize and understand the relationship between sensors’ performance and other factors (or sensors).

• E.g., Condenser Water Pumps vs time.

Page 14: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

14

Summary• Development of software tools to facilitate the process of collecting

data from NWSC, performing basic quality control, and analyzing as well as visualizing key parameters from data collected.

Page 15: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

15

Future Work

• Two possibilities:– Configuring sMAP properly and develop a high level custom

application on top of sMAP archiver for NWSC staff to monitor the facility.

Page 16: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

16

Future Work

• Two possibilities:– Stick with CSV files and take advantage of Google Fusion

Tables, an experimental yet powerful data visualization web application developed by Google. [2]

Page 17: 1 Optimizing the Efficiency of the NCAR-Wyoming Supercomputing Center Facility A Software Perspective Theophile Nsengimana Collaborator: Ademola Olarinde

17

References

[1] sMAP 2.0 documentation http://www.cs.berkeley.edu/~stevedh/smap2/

[2] Google Fusion Table Help Center https://support.google.com/fusiontables/?hl=en