Plugfest Learnings_r4.pptx

18
Learnings from the first Plugfest Tetsuya Nakamura, CableLabs

Transcript of Plugfest Learnings_r4.pptx

Learnings from the first Plugfest

Tetsuya Nakamura, CableLabs

OPNFV Plugfest overview

• May 9 – 13, 2016 @ CableLabs (Louisville, CO, USA)• Objective

• Provide an opportunity to test and demonstrate interoperability of NFV solutions highlighting the OPNFV platform

• Key areas of testing

OPNFV deployment

Integration with different hardware and installer (Apex, Compass, Fuel, JOID) platforms

Network integration

Integration of OPNFV platform with SDN Controller or other networking hardware

VNF deployment

Verification of VNF life cycle on OPNFV platform(s), including deployments created by different installers

Planning

• Bi-weekly meetings• Mailing list for Plugfest participants• Planning wiki for participants

• Attendee info• Testing goals & plans• Hardware/software resources on site• Testing schedule

• Participants were required to accept the Rules of Engagement at registration

• Plugfest is NOT a certification process• 60-day “Privacy Period” following the Plugfest• Plugfest results are not to be used for comparative marketing purposes

Participants

• 41 participants from 19 organizations (incl. 3 non-members)• Had representatives from all installer projects and key testing projects• Hardware resources were available on-site (from Huawei & Intel) and via Pharos Labs

What went well…

• Face-to-face collaboration• Break out sessions

• Testing projects (Functest, Yardstick, and StorPerf)• SDN controller integration with OPNFV installers

• Daily meet ups to recap progress and discuss next steps• On-site hardware

Release: June, 22nd

Huawei on-site Pharos PODs

• Huawei E9000• 384GB RAM• 800GB SSD• 4x10Gbps +

2x40Gbps NICs

• The rack containing the 2 Pharos pods• 14 servers

Intel on-site Pharos PODs

• Six Intel® Xeon® based “white-box”• Intel® Server Board “Wildcat Pass”• Intel® Xeon® E5-2699 v4 LGA2011

2.2GHz 55MB 145W 22 cores• 64GB RAM• 480GB SSD 2.5in SATA 3.0 6Gb/s Intel®

“Haleyville”• 3TB SATA Hard Disk Drive• Intel® 10 Gigabit Server Adapters

•The rack containing the 3 Pharos pods• 18 servers

Integration of functional and performance testing

• Functest• Base system functionality, component

testing• Yardstick

• End to end system level performance testing

• StorPerf• Storage performance

• Cperf• Controller performance

Functest

• 62 test results in total• 5 different systems on 3 different

hardware• 3 scenarios

• vPing (SSH, userdata)• Tempest• Rally bench• ONOS• Promise• vIMS

Yardstick• 4 different systems on 3 different

hardware• 2 scenarios• 3 installers (Compass, Fuel, commercial)

• Network latency with Ping• Memory and storage latency• Variations in packet delay (jitter)• Memory read and write bandwidth• Processing speed• Networking throughput and packet loss

StorPerf

• Queue depths• 1, 4, 8

• Block sizes• 4k, 16k

• Volume size• 100 GB

• Workload• Mix of 70% Read

/ 30% Write

Test scenario: E2E interoperability

• Compass installer • 50 integration test

items • VNF lifecycle• Different platform

Test scenario: vEPC

Test scenario: Others

• Software-Defined scalable storage• A freeware storage subsystem integration into 3 installers: Apex, Fuel, JOID

• vIMS on multiple hardware• Armband testing of vIMS

• Automated deployment steps that would be normally invoked in Functest by an independent cloud orchestrator.

• 2 installers (Apex, JOID) for ARM-based remote Pharos PODs

• OpenDaylight with new installers and test tools• 2 installers: Fuel, JOID• Cperf testing

Summary of major findingsKey Testing Area Accomplishments

OPNFV Deployments: hardware, platforms, installers

• Additional installer support was provided for new platforms• Upstream enhancements to installers were recommended and planned• Alternate storage subsystems were installed and validated on all four major

installers (Apex, Compass, Fuel and JOID), including their interaction with OpenStack, and upstream needs were also identified

• Semiconductor vendors installed and tested OPNFV on their own “white boxes” to gather usability information. All major installers and testing projects were utilized for this effort

Network Integration: testing scenarios for components in different configurations

• Functional and performance tests were integrated in valuable ways• Yardstick measurements were run on multiple platforms, installers and applications• Controller integration was achieved for both OpenDaylight and ONOS. For ODL,

upstream work was started for Fuel and JOID, and CPerf testing was evolved. ONOS was installed with Fuel using Ubuntu as the host OS, and was tested with both Functest and Yardstick

VNF Applications in new environments

• A vIMS and a vEPC system were successfully deployed on new combinations of hardware and software

• Many VNFs were tested on all four installers. • A virtual performance and analytics was also installed and tested in a multivendor

environment.

E2E Interoperability • A collaborative team provided E2E integration including COTS hardware, VNFM, VNF• The team also used the Compass installer tool, and contributed 50 integration test

cases to OPNFV based on discoveries in testing NFV reference points

Participating OPNFV member organizations

Thoughts for the next OPNFV Plugfest

• More on-site hardware resources• e.g. dedicated hardware for each installer team

• Divide the Plugfest into two parts• Part I: Focused on install tools & testing• Part II: Focused on interoperability (e.g. for VNF vendors)

• Would like to see more participation from VNF and network element vendors

• Breakout session for the next Plugfest planning• Date: Today(June,20th) at 14:40 – 15:40• Room: Bellevue

THANK YOU

18

THANK YOU