london openstack meetup: measuring performance: the basics

15
Measuring Performance: the basics by Stack Evolution

Upload: darryl-weaver

Post on 07-Feb-2017

37 views

Category:

Internet


1 download

TRANSCRIPT

Page 1: London Openstack Meetup: Measuring performance: the basics

Measuring Performance: the basicsby Stack Evolution

Page 2: London Openstack Meetup: Measuring performance: the basics

2 © Stack Evolution Ltd

Why?I just finished building my IaaS cloud.

But…

Is it any good?

Page 3: London Openstack Meetup: Measuring performance: the basics

3 © Stack Evolution Ltd

Lots of metrics and measuresSystem monitoringSystem metric graphingLog file analysis

These are focused on the infrastructure, not the tenant workload

Page 4: London Openstack Meetup: Measuring performance: the basics

4 © Stack Evolution Ltd

IaaS is a Shared PlatformTunable memory/processor/disk overcommitDynamic workloadsBatch jobsTime of day/month/yearMany variables

Page 5: London Openstack Meetup: Measuring performance: the basics

5 © Stack Evolution Ltd

Measuring the tenant workloadsTenant only cares about workload

performanceHow does the disk perform?Is the processor faster or slower?What is performance of a specific Database?How do HPC jobs perform?

Page 6: London Openstack Meetup: Measuring performance: the basics

6 © Stack Evolution Ltd

BenchmarksMany standard benchmark tools available:IperfUnixbenchFioSysbench_OLTPAerospikeRedisScimark2HPCC

Page 7: London Openstack Meetup: Measuring performance: the basics

7 © Stack Evolution Ltd

Standardise Suite of BenchmarksSelect flavours to directly compare memoryVolumes of standard size (20GB)Network: Internal and ExternalSpecific applications

DatabaseLoggingStreamingHPC

Page 8: London Openstack Meetup: Measuring performance: the basics

8 © Stack Evolution Ltd

Run on multiple platformsCollect data from

Public CloudsPrivate cloud deploymentsStaging/Test Clouds

Page 9: London Openstack Meetup: Measuring performance: the basics

9 © Stack Evolution Ltd

CompareNow compare the data to evaluate how good your cloud really is.

Page 10: London Openstack Meetup: Measuring performance: the basics

Demo

Page 11: London Openstack Meetup: Measuring performance: the basics

11 © Stack Evolution Ltd

Variability of performance for similar flavours across cloud providers, notice the latency and boot time graphs higher is worse, others higher is better

Page 12: London Openstack Meetup: Measuring performance: the basics

12 © Stack Evolution Ltd

Overall Index score for combined benchmarks, notice the variability in providers is as much as 7 times as much performance for the same flavour size

Page 13: London Openstack Meetup: Measuring performance: the basics

13 © Stack Evolution Ltd

Mimicking real world workloads: Database and Logging and measuring disk performance using a 20GB Volume of a standard type

Page 14: London Openstack Meetup: Measuring performance: the basics

14 © Stack Evolution Ltd

A High Performance Computing Benchmark, then weighted by flavour hourly cost and finally weighted by Volume monthly cost (for 20GB Volume).

Notice how much the graphs change once you add in pricing information to the performance data.

Page 15: London Openstack Meetup: Measuring performance: the basics

15 © Stack Evolution Ltd

Data provided by:

Thanks

Contact [email protected] if you want to find out more.