load prediction for best effort real time
DESCRIPTION
CMU load prediction software uses linear time series models to predict load on each host. QuO delegate choses server replica based on load predictions Integration by Peter A. Dinda and Xiaoming Liu. Application code. ARFIMA(p,d,q). Client. BBN QuO System. Infinite # roots. - PowerPoint PPT PresentationTRANSCRIPT
Load Prediction for Best Effort Real TimePeter A. Dinda
[tmin,tmax] ??InteractiveApplication
Short taskswith deadlines
Unmodified COTS Distributed System
1 3 5 7Measured Load
0
5
10
15
20
25
Exe
cutio
n T
Ime
(Sec
onds
)
42,000 pointsCoefficient of Correlation = 0.998
nominal
tt
t
tdttload
execnow
now
)(1
1
0
0.2
0.4
0.6
0.8
1
1.2
Host
Production Cluster ResearchCluster
Desktops
+SDev
-SDev
Mean
Title:axp7_tue_19.epsCreator:MATLAB, The Mathworks, Inc.Preview:This EPS picture was not savedwith a preview included in it.Comment:This EPS picture will print to aPostScript printer, but not toother types of printers.
Title:axp7_19_day_time.epsCreator:MATLAB, The Mathworks, Inc.Preview:This EPS picture was not savedwith a preview included in it.Comment:This EPS picture will print to aPostScript printer, but not toother types of printers.
Time
0 20000 40000 60000 80000
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
tj
jtjt aaz
1
Time
0 20000 40000 60000 80000
-0.0
4-0
.02
0.0
0.0
20
.04
),0(~ 2at WhiteNoisea 2,~ ztz
22za
UnpredictableRandom Sequence Fixed Linear Filter
Partially PredictableLoad Sequence
ARFIMA(p,d,q)
ARIMA(p,d,q)
ARMA(p,q)
AR(p) MA(q)
tt aB
z)(
1
tt aB
Bz
)(
)(
tt aBz )(
0,integerfor)1)((
)(
dda
BB
Bz tdt
5.00for)1)((
)(
da
BB
Bz tdt
Infinite # roots
d roots
a is the confidence interval for t+1 predictions
Map work that would take 100 ms at zero load
axp0: z=0.54, =1.0, a(ARMA(4,4))= 0.109 a(ARFIMA(4,d,4))= 0.108no model: 1.0 +/- 1.06 (95%) => 100 to 306 msARMA: 1.0 +/- 0.22 (95%) => 178 to 222 msARFIMA: 1.0 +/- 0.21 (95%) => 179 to 221 ms
axp7: z=0.14, =0.12, a(ARMA(4,4))= 0.041 a(ARFIMA(4,d,4))= 0.025no model: 0.12 +/- 0.27 (95%) => 100 to 139 msARMA: 0.12 +/- 0.08 (95%) => 104 to 120 msARFIMA: 0.12 +/- 0.05 (95%) => 107 to 117 ms
Delegatewith Hysteresis
LoadPredSyscond
Load Prediction
Engine
Host A
OtherJobs
LoadPredSyscond
ServerReplica
ServerReplica
Load Prediction
Engine
Host B
OtherJobs
ClientApplication code
BBN QuO System
CMU Load Prediction(Peter A. Dinda)
Delegate choosesserver replica basedon load predictions
•CMU load prediction software uses linear time series models to predict load on each host.•QuO delegate choses server replica based on load predictions
Integration by Peter A. Dinda and Xiaoming Liu
Execution Model Execution Time and Host Load Integration In BBN QuO
Load is Self Similar Load Exhibits Epochal Behavior
Linear Time Series Models Realizable Pole-zero Models Real World Benefits of Prediction
ARFIMA ModelsCapture Long-rangeDependence of Self-Similar Signals
Choose host based on predicted load
-1
-0.5
0
0.5
1
1.5
2
Host
+SDev
-SDev
Mean
Production Cluster ResearchCluster
Desktops
Load is Highly Variable
Execution Modeland Integrationinto BBN QuO
StatisticalProperties ofHost Load
Load PredictionWith Linear Time
Series Models