Misalignments To Reduce Resource Over-Provisioning in Cloud - - PowerPoint PPT Presentation

misalignments to reduce
SMART_READER_LITE
LIVE PREVIEW

Misalignments To Reduce Resource Over-Provisioning in Cloud - - PowerPoint PPT Presentation

Considering Resource Demand Misalignments To Reduce Resource Over-Provisioning in Cloud Datacenters Liuhua hua Chen en Haiyi ying ng Shen en Dept pt. of E f Elec ectric trical al and d Co Compu puter ter Eng. g. Dept pt. of Co


slide-1
SLIDE 1

Considering Resource Demand Misalignments To Reduce Resource Over-Provisioning in Cloud Datacenters Haiyi ying ng Shen en

Dept

  • pt. of Co

f Compu pute ter r Sc Science ence University ersity of f Vi Virgi ginia, nia, USA SA

Liuhua hua Chen en

Dept

  • pt. of E

f Elec ectric trical al and d Co Compu puter ter Eng. g. Cl Clems mson

  • n University,

ersity, USA SA

slide-2
SLIDE 2

Cloud Computing

  • Prof. Haiying Shen, University of Virginia

2

  • Cloud computing: large groups of remote servers networked to

allow centralized data storage and online access to computer services or resources

slide-3
SLIDE 3

Cloud Providers

  • Prof. Haiying Shen, University of Virginia

3

slide-4
SLIDE 4

Cloud Customers

  • Prof. Haiying Shen, University of Virginia

4

slide-5
SLIDE 5

Research Problem and Goal

5

  • Prof. Haiying Shen, University of Virginia

Virtual machines (VMs) Physical machines (PMs)

slide-6
SLIDE 6

Motivation

  • Prof. Haiying Shen, University of Virginia

6

  • Over-loaded PMslow QoSSLO violationpenalty
  • Under-loaded PMs resource wastehigh system cost
  • Problem: reduce over-loaded and under-loaded PMs
  • Goal: high QoS, high resource efficiency, high profit

PM VM VM VM VM VM VM VM PM VM PM VM VM VM Virtual machine (VM) VM VM VM Physical machine (PM)

slide-7
SLIDE 7

PM

Initial Complementary VM Consolidation

  • Previous work (CompVM):

VM VM VM VM PM PM PM PM

  • How to achieve ?

VM VM VM VM

  • Prof. Haiying Shen, University of Virginia

7

  • L. Chen and H. Shen, Consolidating Complementary VMs with Spatial/Temporal-

awareness in Cloud Datacenters, Proc. of the 33rd Annual IEEE International Conference on Computer Communications (INFOCOM'14), Toronto, Canada, 2014

slide-8
SLIDE 8

CPU utilization (%)

VM1 VM2

100 100 Memory utilization (%) Time Utilization (%) T 100

VM1 VM2 VM3

Spatial Temporal

High CPU low MEM Low CPU high MEM

Initial Complementary VM Consolidation – Motivation

  • CompVM: load balance in the long term

consolidate complementary VMs

VM2 VM1 VM1 VM1

Patterns?

Demand Capacity

Resource waste Under-loaded PMs

Demand Capacity

Over-loaded PMs

Demand Capacity

No under/over-loaded PMs

  • Prof. Haiying Shen, University of Virginia

8

slide-9
SLIDE 9

Initial Complementary VM Consolidation – VM Utilization Pattern

  • Measurement:

– MapReduce jobs: TeraSort, TestDFSIO read/write – cluster trace, trace

  • Periodic resource utilization patterns exist in many VMs running

– the same short-term job – a long-term job

TestDFSIO read Google cluster trace TestDFSIO read

  • Prof. Haiying Shen, University of Virginia

9

slide-10
SLIDE 10

Initial Complementary VM Consolidation – Utilization Pattern Detection

0.0 1.0 2.0 3.0 4.0 5.0 12 24 36 48 60 72

Time (hr) CPU utilization (%)

trace 2 trace 1 pattern f3 f1 f2 u

  • Prof. Haiying Shen, University of Virginia

10

slide-11
SLIDE 11

Initial Complementary VM Consolidation – VM Allocation Method

For each PM: check if it has sufficient capacity for the VM Select the PM: with the least remaining resource after allocating the VM Detect resource utilization pattern of the VM

𝐹

𝑘: ratio of aggregated resource demand

𝑁

𝑘: distance between the average resource

demand vector and the capacity vector

pattern

Memory resource waste

PM VM4 VM5 VM6 PM VM2 PM VM1 VM3 VM7 PM1 PM2 PM3 Choose a PM PM VM4 VM5 VM6 PM VM2 PM VM1 VM3 VM7 PM1 PM2 PM3 Choose a PM For each PM: check if it has sufficient capacity for the VM Select the PM: with the least remaining resource after allocating the VM Detect resource utilization pattern of the VM

consolidate complementary VMs

  • Prof. Haiying Shen, University of Virginia

11

PM VM4 VM5 VM6 PM VM2 PM VM1 VM3 VM7 PM1 PM2 PM3 Choose a PM

CPU utilization (%)

VM1 VM2

100 100 Memory utilization (%) Time Utilization (%) T 100

VM1 VM2 VM3

Spatial Temporal

High CPU low MEM Low CPU high MEM

slide-12
SLIDE 12

Reducing Prediction Error

12

  • Prof. Haiying Shen, University of Virginia

pattern

slide-13
SLIDE 13

VM Consolidation

– Reduce Provisioned Resource

  • Pulse deviation yields a pattern with a pulse width

larger than the actual pulse width

  • Resource over-provisioning
  • Not revealed or studied before

13

  • Prof. Haiying Shen, University of Virginia

deviation derived pattern derived pattern Pulse

slide-14
SLIDE 14

VM Consolidation

– Trace Study

  • Pulse deviations are common

14

  • Prof. Haiying Shen, University of Virginia

Google Cluster trace PlanetLab trace

100 jobs, 29920 tasks 1000 jobs, 4695 tasks

Average task execution time is around 100 minutes/seconds MapReduce benchmarks on a HPC cluster (Wordcount, Grep, Terasort, TestDFSIO and PiEstimator)

slide-15
SLIDE 15

VM Consolidation

– Trace Study

  • Resource efficiency: demand/capacity
  • Even using CompVM, the resource efficiency still

needs to improve

15

  • Prof. Haiying Shen, University of Virginia

Google Cluster trace PlanetLab trace

100 jobs, 1550 tasks 1000 jobs, 4695 tasks

slide-16
SLIDE 16

VM Consolidation

– Pattern Refinement Methodology

16

  • Prof. Haiying Shen, University of Virginia
  • Pattern refinement methods

– Lowering cap – Reducing pulse width – Optimal base provisioning

Time unused resource t1 t2 t3 chigh clow

: lower each value in the pattern by chigh-clow

slide-17
SLIDE 17

VM Consolidation

– Pattern Refinement Methodology

17

  • Prof. Haiying Shen, University of Virginia
  • Pattern refinement methods

– Lowering cap – Reducing pulse width – Optimal base provisioning : postpone the pulse from t1 to t3

slide-18
SLIDE 18

VM Consolidation

– Pattern Refinement Methodology

18

  • Prof. Haiying Shen, University of Virginia
  • Pattern refinement methods

– Lowering cap – Reducing pulse width – Optimal base provisioning

Time b1 b2

refine pattern based on optimal b value that maximizes resource efficiency

slide-19
SLIDE 19

VM Consolidation

– Pattern Refinement Methodology

19

  • Prof. Haiying Shen, University of Virginia
  • Pattern refinement methods

– Lowering cap – Reducing pulse width – Optimal base provisioning

  • Risk violating SLOs
slide-20
SLIDE 20

20

  • Prof. Haiying Shen, University of Virginia

Cluster experiment Google Cluster trace

Pattern refinement yields higher resource efficiency without compromising VM performance by handling pulse deviations!

VM Consolidation

– Performance Evaluation

2000 VMs, CloudSim, Palmetto HPC cluster, Traces: NAS Parallel Benchmark, cluster

Improve by 10%, 70% Improve by 40%

slide-21
SLIDE 21

Conclusion

  • Trace study

– Pulse deviations are common – Even using CompVM, the resource efficiency still needs to improve

  • Pattern refinement methods

– Lowering cap – Reducing pulse width – Optimal base provisioning

  • Experiments

– Higher resource efficiency without compromising VM performance

21

  • Prof. Haiying Shen, University of Virginia
slide-22
SLIDE 22

Future Work

  • Consider other factors (e.g., SLOs) in VM

consolidation

  • Consider VM migration

22

  • Prof. Haiying Shen, University of Virginia
slide-23
SLIDE 23

23

  • Prof. Haiying Shen, University of Virginia

Thank you! Questions & Comments?

Haiying Shen

hs6ms@virginia.edu Associate Professor Department of Computer Science University of Virginia