online feedback-based estimation of dynamic page service time

20
Online Feedback-based Estimation of Dynamic Page Service Time Ashwini Kumar Kaushik Veeraraghavan Benjamin Wester Kang Shin

Upload: ember

Post on 31-Jan-2016

30 views

Category:

Documents


0 download

DESCRIPTION

Online Feedback-based Estimation of Dynamic Page Service Time. Ashwini Kumar Kaushik Veeraraghavan Benjamin Wester Kang Shin. Motivation. Dynamic pages Increasingly prevalent Pages difficult to cache Service time applications Differentiated QoS (prioritizing requests) Load balancing - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Online Feedback-based Estimation of Dynamic Page Service Time

Online Feedback-based Estimation of Dynamic Page

Service Time

Ashwini KumarKaushik Veeraraghavan

Benjamin WesterKang Shin

Page 2: Online Feedback-based Estimation of Dynamic Page Service Time

2

Motivation

• Dynamic pages– Increasingly prevalent– Pages difficult to cache

• Service time applications– Differentiated QoS (prioritizing requests) – Load balancing– Virtual host services– Estimates: credit card payment times

Page 3: Online Feedback-based Estimation of Dynamic Page Service Time

3

Idea

• History– Maintain records of past service times– Predict based on existing records

• Assume a correlation between past and future– What is the invariant property?– What system state can we store to track it?

• We explore three estimators and their design

Page 4: Online Feedback-based Estimation of Dynamic Page Service Time

4

Outline• Motivation

• Design & Implementation

• Evaluation

• Conclusion and Future Work

Page 5: Online Feedback-based Estimation of Dynamic Page Service Time

5

History

1. Maintain per-URL fixed-size history table2. Data depends on the estimator used3. Tag incoming requests4. Update history when request processing finishes

Page 6: Online Feedback-based Estimation of Dynamic Page Service Time

6

SEDA: Haboob

Page 7: Online Feedback-based Estimation of Dynamic Page Service Time

7

Sirocco

Page 8: Online Feedback-based Estimation of Dynamic Page Service Time

8

Estimators

Scheme Store in history Estimation

Average

STQ

STQT

Length Queue

Time Service

Time Service

Length Queue

Threads Time Service

)(historyavg

Length Queuehistoryavg )(

Threads

Length Queuehistoryavg )(

Page 9: Online Feedback-based Estimation of Dynamic Page Service Time

9

Outline• Motivation

• Design & Implementation

• Evaluation

• Conclusion and Future Work

Page 10: Online Feedback-based Estimation of Dynamic Page Service Time

10

Evaluation

• How close are our estimators?

• Load patterns– Steady load: httperf

• 50ms / request

– Changing load: SURGE

0

20

40

60

80

100

0 5 10 15 20 25

Re

que

sts

Time (seconds)

Page 11: Online Feedback-based Estimation of Dynamic Page Service Time

11

Average Estimator (steady)

• Request takes ~35ms to complete

• Estimator should not be too responsive

Page 12: Online Feedback-based Estimation of Dynamic Page Service Time

12

Average Estimator (changing)

• Each point averages 20 requests

• Vertical bars indicate the addition of a handling thread

Page 13: Online Feedback-based Estimation of Dynamic Page Service Time

13

Average Estimator

• How to read the graph:• Enters system: t = 5.4s• Estimated service time: 1.6s• Real service time: 4.9s

Page 14: Online Feedback-based Estimation of Dynamic Page Service Time

14

Average Estimator

• History becomes out-of-date

• Base estimation on the current state of the system

Page 15: Online Feedback-based Estimation of Dynamic Page Service Time

15

STQ Estimator

• It’s better than the Average Estimator…

• Inaccurate around a thread change

Page 16: Online Feedback-based Estimation of Dynamic Page Service Time

16

STQ Estimator

• Estimations will be inaccurate unless the estimator knows about the coming change in the number of threads

• After a change, recently-completed requests show a mixed view of system state and will pollute the history

Page 17: Online Feedback-based Estimation of Dynamic Page Service Time

17

STQT Estimator

• Overshoots target because of a polluted history

• Fix: Integrate request times across the thread change

Page 18: Online Feedback-based Estimation of Dynamic Page Service Time

18

Outline• Motivation

• Design & Implementation

• Evaluation

• Conclusion and Future Work

Page 19: Online Feedback-based Estimation of Dynamic Page Service Time

19

Concluding Remarks

• Takeaways– Don’t adapt to transient delays– Use current state along with the history– Be aware of underlying changes

• Future Directions– Extrapolate requests’ service times– URL service time as a distribution

Page 20: Online Feedback-based Estimation of Dynamic Page Service Time

Questions?