Network Bandwidth Utilization Forecast Model on High Bandwidth - - PowerPoint PPT Presentation

network bandwidth utilization forecast model on high
SMART_READER_LITE
LIVE PREVIEW

Network Bandwidth Utilization Forecast Model on High Bandwidth - - PowerPoint PPT Presentation

Network Bandwidth Utilization Forecast Model on High Bandwidth Networks Scientific Data Management Group Computational Research Division Lawrence Berkeley National Laboratory Wucherl (William) Yoo, Alex Sim Feb. 17, 2015 W. Yoo, CRD , LBNL 1


slide-1
SLIDE 1
  • W. Yoo, CRD, LBNL

1

  • Feb. 17, 2015

Network Bandwidth Utilization Forecast Model on High Bandwidth Networks

Scientific Data Management Group Computational Research Division Lawrence Berkeley National Laboratory Wucherl (William) Yoo, Alex Sim

slide-2
SLIDE 2
  • W. Yoo, CRD, LBNL

2

  • Feb. 17, 2015

Motivation

  • Increasing Data Volume
  • Efficient resource management and scheduling data

movement

  • Predict the network bandwidth utilization between two HPC sites
  • Challenge
  • Accurate and fine-grained performance model
  • Computational complexities and variances/burstiness
slide-3
SLIDE 3
  • W. Yoo, CRD, LBNL

3

  • Feb. 17, 2015

SNMP Data

  • Simple Network Management Protocol (SNMP) Data
  • Collected by ESNet in 2013 and 2014 on each router
  • Connect a pair of large data facilities
  • P1 and P2 between NERSC and ORNL
  • P3 and P4 between NERSC and ANL
  • P5 and P6 between ORNL and ANL
  • Univariate time series with bandwidth utilization size

and time-scale at 30s interval

slide-4
SLIDE 4
  • W. Yoo, CRD, LBNL

4

  • Feb. 17, 2015

Bandwidth Utilization

NERSC è ANL (P1) ANL è NERSC (P2) NERSC è ORNL (P3) ORNL è NERSC (P4) ANL è ORNL (P5) ORNL è ANL (P6)

slide-5
SLIDE 5
  • W. Yoo, CRD, LBNL

5

  • Feb. 17, 2015

Our Approach

  • Seasonal Adjustment
  • Logit Transformation
  • Stationarity
  • Delayed Model Update
slide-6
SLIDE 6
  • W. Yoo, CRD, LBNL

6

  • Feb. 17, 2015

Prediction Model

  • Forecast Error
  • Logit Transform
  • lower bound a, upper bound b
  • Prediction Models
  • ARIMA, ETS, and Random Walk
slide-7
SLIDE 7
  • W. Yoo, CRD, LBNL

7

  • Feb. 17, 2015

Seasonal Adjustment

  • STL
  • A sequence of smoothing from Loess (Locally Weighted

Regression Fitting)

  • Decomposes the logit transformed SNMP time series into

the components S, T, and R.

  • Seasonality S
  • Trend T
  • Remainder R
slide-8
SLIDE 8
  • W. Yoo, CRD, LBNL

8

  • Feb. 17, 2015

Stationarity

  • Stationarity
  • The mean or variance of time-series does not change over

time and does not follow any trends

  • Burstiness
  • When there is a sudden bandwidth utilization change, the

time series can be looked as non-stationary

  • Keeping the stationary assumption made less prediction

error in our model

  • bursty change may not be a long-term change
slide-9
SLIDE 9
  • W. Yoo, CRD, LBNL

9

  • Feb. 17, 2015

Delayed Model Update

  • Based on the stationarity, keeping the same model

and delaying model updates

  • Instead of refitting, the minimal parts such as auto-

correlation and moving averages are updated from the initially fitted ARIMA model

slide-10
SLIDE 10
  • W. Yoo, CRD, LBNL

10

  • Feb. 17, 2015

Evaluation

  • Forecast Model Comparison
  • Logit Transformation
  • Training Set Size
  • Stationarity
  • Delayed Model Update
  • Forecast Results
slide-11
SLIDE 11
  • W. Yoo, CRD, LBNL

11

  • Feb. 17, 2015

Forecast Model Comparison

0e+00 1e+08 2e+08 3e+08 4e+08 5e+08 P1 P2 P3 P4 P5 P6

Path RMSE

Model ARIMA ETS RW

slide-12
SLIDE 12
  • W. Yoo, CRD, LBNL

12

  • Feb. 17, 2015

Logit Transformation

0e+00 1e+08 2e+08 3e+08 4e+08 P1 P2 P3 P4 P5 P6

Path RMSE

Data Logit Transformed Unmodified

slide-13
SLIDE 13
  • W. Yoo, CRD, LBNL

13

  • Feb. 17, 2015

Training Set Size

0e+00 1e+08 2e+08 3e+08 4e+08 P1 P2 P3 P4 P5 P6

Path RMSE

TrainingWeeks 1 2 3 4 5 6 7 8

slide-14
SLIDE 14
  • W. Yoo, CRD, LBNL

14

  • Feb. 17, 2015

Stationarity

0e+00 1e+08 2e+08 3e+08 P1 P2 P3 P4 P5 P6

Path RMSE

Data Non−stationary Stationary

slide-15
SLIDE 15
  • W. Yoo, CRD, LBNL

15

  • Feb. 17, 2015

Forecast Results

NERSC è ANL ANL è NERSC NERSC è ORNL ORNL è NERSC ANL è ORNL ORNL è ANL

slide-16
SLIDE 16
  • W. Yoo, CRD, LBNL

16

  • Feb. 17, 2015

Historical Forecast Results

slide-17
SLIDE 17
  • W. Yoo, CRD, LBNL

17

  • Feb. 17, 2015

Forecast Results - RMSE

PID SD_Train SD_Test RMSE

P1 4.13 2.36 2.27 P2 4.51 3.37 3.31 P3 4.01 2.07 1.88 P4 3.03 2.06 1.85 P5 4.64 3.40 3.42 P6 4.00 2.54 2.42

slide-18
SLIDE 18
  • W. Yoo, CRD, LBNL

18

  • Feb. 17, 2015

Conclusion

  • Forecast Model
  • ARIMA with STL, logit transformation, and stationarity
  • Forecast errors were within the variances of observed data
  • Logit transform reduced prediction error by 8.5%
  • Stationarity assumption reduced prediction error by 10.9%
  • Contact
  • Wucherl (William) Yoo, wyoo@lbl.gov
slide-19
SLIDE 19
  • W. Yoo, CRD, LBNL

19

  • Feb. 17, 2015

Backup - Future Work

  • Adaptive Model
  • To adapt the long-term trend changes
  • Multivariate Performance Prediction Model
  • To extend the analysis to multivariate data
slide-20
SLIDE 20
  • W. Yoo, CRD, LBNL

20

  • Feb. 17, 2015

Backup - Seasonal Adjustment