Transcript
Page 1: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential Cisco Confidential © 2013 Cisco and/or its affiliates. All rights reserved.

Predictive Analytics & Visualization Towards Data Driven Insights for Openstack Clouds

Page 2: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential Cisco Confidential Cisco Confidential © 2013 Cisco and/or its affiliates. All rights reserved.

Cloud/Big-data Winnovation Team

Alex Holden Damian Van Vuuren Ailing Zhang

Marc Solanas Matt Weeks Pengfei Zhang

Kai Zhang

Xinyuan Huang

Debo Dutta

Yathiraj Udupi

Page 3: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Visualization in OpenStack today

Page 4: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Tomorrow?

Sources: 2001 A Space Oddesy, MGM. Minority Report, Twentieth Century Fox Film Corporation. Mad Men, AMC

OpenStack, deploy me an

app

Page 5: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Our Approach: Predictive Analytics & Awesome Visualisation Why Predictive? ●  OpenStack has lots of moving parts ●  Rule/Model/Policy based system hard without insight

Why Predictive? ●  Ops folks want to know right now!

Why Visualisation? ●  Easy to observe insights by humans ●  East way to consume lots of information from loads oflogs/metrics ●  Improve user experience

Page 6: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential Cisco Confidential © 2013 Cisco and/or its affiliates. All rights reserved.

Digital Exhaust = Metrics/Logs What information can we get from our cloud? Where can we get it?

●  Ceilometer ●  OpenStack Syslog ●  Project APIs (Nova, Neutron, Cinder etc)

What can we do with it? ●  Extract Insights ●  Figure out what failed and why ●  Move workloads around ●  Detect attacks on the cloud

●  Physical Servers ○  CPU ○  Hardware Faults ○  OpenStack Logs ○  OpenStack Messaging Queue

●  Tenant/Project Data ○  Quotas

●  Virtual Machines ○  Config ○  CPU ○  Disk ○  Network Flows ○  Performance ○  Latency

●  Application Level ○  Application Logs ○  Application Requirements ○  Application Performance

Page 7: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential Cisco Confidential Cisco Confidential © 2013 Cisco and/or its affiliates. All rights reserved.

AVOS Dashboard

Page 8: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

AVOS: Summary ●  Analytics and Visualization on OpenStack ●  Stateless: easy deployment ●  Client side architecture:

○  Polls some OpenStack APIs ○  Listens to the messaging bus ○  Minimum overhead

●  Provides users and service providers full insight into cloud

○  Real time info on all instances/networks etc by project and as admin ○  Historical data on performance ○  Stack-search: Search for all components in the cloud, quickly. ○  Multiple select (See combined data)

Page 9: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

https://www.youtube.com/watch?v=vyb-Cd4VEDg

AVOS Demo

Page 10: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Bottlenecks

Use Cases: Why use AVOS?

Attacks

Page 11: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Use Case: Bottlenecks

Page 12: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Use Case: Attack Detection ●  Extract Metrics ●  Analyse Patterns ●  We can differentiate

between: ○  Hadoop Traffic ○  Network Faults ○  DDoS Attack

Page 13: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Future Plans •  Advanced Search Primitives •  Reducing the volume of logs •  Real-time error analytics •  Add to/edit the cluster •  Ability to preview changes

Curvature:

https://github.com/CiscoSystems/curvature

Page 14: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Notes: OpenStack changes •  Ceilometer

Better API queries, optimizing return structure Custom meters: VM to VM traffic flows

•  Other suggestions/Ideas

Standardization of the Python clients (several inconsistencies between method calls) Integration into horizon

Page 15: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Conclusion

Openstack Predictive Analytics/Visibility = Reduce Ops pains We must 1) gather data 2) extract insights 3) present to ops folks with awesome UX

Page 16: Predictive Analytics & Visualization - Towards Data Driven Insights for Openstack Clouds

© 2013 Cisco and/or its affiliates. All rights reserved. Cisco

Confidential

Thank you. Please take our survey: http://goo.gl/CMAZje

Coming Talk: Storage Visibility and Optimization for OpenStack:

Wednesday, May 14 • 3:30pm - 4:10pm


Recommended