kddi - openstack summit 2016/red hat nfv mini summit

18
KDDI Research Inc. Proprietary and Confidential Troubles prediction and detection based on Distributed Monitoring & Analytics framework Yuki Kasuya <[email protected]> KDDI Research

Upload: kimw001

Post on 15-Apr-2017

183 views

Category:

Technology


0 download

TRANSCRIPT

KDDI  Research  Inc.  Proprietary  and  Confidential

Troubles  prediction  and  detection  based  on  Distributed Monitoring  &  Analytics  frameworkYuki  Kasuya  <yu-­kasuya@kddi-­research.jp>  KDDI  Research

KDDI  Research  Inc.  Proprietary  and  Confidential

Agenda

1

2

3

4

Motivation  /  Problem

Solution  /  Architecture

Use  case

Conclusion

2

KDDI  Research  Inc.  Proprietary  and  Confidential

1.  Motivation  /  Problem

KDDI  Research  Inc.  Proprietary  and  Confidential

RemoteOperation  Center

Development  Division

nRequire  reliability

nMany  operators  needed

nHigh  cost  operationl change  to  low  cost

4

Current  operation  style

Data  Center24hours  /  365daysHW  replacement

daytimesoftware  bug

KDDI  Research  Inc.  Proprietary  and  Confidential

Driver:  Changing  operation  style

ReactiveOperation

PeriodicOperation

ProactiveOperation

Cost  Reduction

Agility

Proactive

Reactive

24/7  maintenance

9am  -‐‑‒ 5pmmaintenance

Automation

5

before  broken,do  prevention

KDDI  Research  Inc.  Proprietary  and  Confidential

What  is  key  point?

n Auto  healing  process

1. Fault  detection  by  monitoring  system

1. Recovery  plan  by  OSS/Orchestrator

1. Auto  healing  by  Orchestrator

For  fast  recovery,  real-­time  fault  detection  /  prediction  is  key  point.

6

KDDI  Research  Inc.  Proprietary  and  Confidential

Problems for  proactive  operation

NFVI NFVI

VMVNFpollerpoller

verbosedata

verbosedata

notifier

evaluator

collector

DBn Centralized  monitoring  architecture

l Difficult  to  real-­time(fine-­grained)  monitoring

l High  load  to  collect  a  lot  of  data

n Generally,  delay  of  collecting  data  affects  several  area.Now,  the  time  has  come  to  consider  to  enhance  the  architecture.

7

delay

KDDI  Research  Inc.  Proprietary  and  Confidential

2.  Solution  /  Architecture

KDDI  Research  Inc.  Proprietary  and  Confidential

n Distribute  each  function  into  computing  nodes

l Monitoring  process  is  complete  in  each  computing  node

l Real-­time(Fine-­grained)  monitoring

l Scale  with  the  number  of  computing  nodes

Distributed  Monitoring  and  Analytics (DMA)

NFVI NFVI

VMVNF

notifier

evaluator

collector

analyzer

evaluator

collector

analyzer

evaluator

collector

DB

DB DB

analyticsresult concise

data

pollerpoller

9

KDDI  Research  Inc.  Proprietary  and  Confidential

Architecture  detail

Poller/Notification

libvirtAPI

SNMP  Get

SNMP  Trap

CollectorDatabase

MeterTranslator

Evaluator

AnalyticsEngineFault

Detection(Prediction)

Statistical  Analysis

Alarm  Correlation

Transmitter

Fault  Alarm

StatisticalData

OpenStackAPI

...

Perf.  Data

Perf.data (CPU/Memory...)

Guest.

(Alarm)Faultdata

Perf.data (CPU/Memory...)

Host.

(Alarm)Faultdata

Fault  Data

...

NFVI

10

KDDI  Research  Inc.  Proprietary  and  Confidential

Centralized Distributed

Fault ✔ ✔(predict,  silent)

Account ✔

Performance✔

(Micro  burst)

Target  area

n Centralizedl cloud  platform

n Distributedl NFVl carrier  grade  network  

11

KDDI  Research  Inc.  Proprietary  and  Confidential

3.  Use  case

KDDI  Research  Inc.  Proprietary  and  Confidential

Use  Case  1:  Prediction  using  machine  learning

T.  Niwa et  al.,  “Universal  Fault  Detection  for  NFV  using  SOM-‐‑‒based  Clustering,”  APNOMS  2015.

Distributed  machine  learning

13

Demo  @MWC2016

KDDI  Research  Inc.  Proprietary  and  Confidential

Use  Case  2:  Detect  micro  burst  traffic

Finer  performance  data  processing

M.  Miyazawa  et  al.,  “In-‐‑‒network  real-‐‑‒time  performance  monitoring  with  distributed  event  processing,”  NOMS  2014.

• Centralized  approach  cannot  detect  the  bursts.• DMA  can  achieve  x1000  finer  monitoring.

-‐‑‒ -‐‑‒   8

-‐‑‒ -‐‑‒   5

-‐‑‒B75 1 B A 1A11 5 8 A5 C1

B AA 1 8

051  A8 5 5 8 7 8   5ADA8 8 1A8 8 55 5

Time%(sec)

0

20

40

60

80

100

0 20 40 60 80 100 120 140

Bandwidth%U

tiliza

tion%(%)

5 A 1 8 5 1 175 5 A

8 A 82BA5 1 175 5 A

C5 H.8 2B A H

70�

18Mbps

DMA  approach

Centralized  approach

14

KDDI  Research  Inc.  Proprietary  and  Confidential

4.  Conclusion

KDDI  Research  Inc.  Proprietary  and  Confidential

nChange  to  proactive  /  low  cost  operation

nFault  detection  /  prediction  is  key  point

nDistributed  Monitoring  and  Analytics  is  suitable  for  proactive  operation

Conclusion

16

KDDI  Research  Inc.  Proprietary  and  Confidential

nStart  to  discuss  how  to  integrate  DMA  to  OpenStack

Join  us

17

KDDI  Research  Inc.  Proprietary  and  Confidential