Do we all have the same vision of Performance ?

End user

...the time to open your application or to visualize your website...

Web Developper

...the number of requests, the bundle size, the page load, ... aka Web Performance Optimization (WPO)...

Back-end Developper

...the time to process requests, the number of requests/sec, scalability, ...

Sysadmins, Ops

...hardware resources (CPU, Memory, IOs) ...

Marketing Guy

...the time of business transaction, the number of business transactions aka Througput...

Manager(s)

... cloud costs, uptime, competitors...

• Who is right ?
• Do we have a problem ?
• Does performance matter ?
• What is performance ?

a common definition

The word performance in computer performance means the same thing that performance means in other contexts, that is, it means ...

How well is the software doing the work it is supposed to do?

How well a software is supposed to work ?

• Short response time for a given action
• High throughput (rate of processing business transaction)
• Low utilization of computing resources(CPU, Memory, IO, ...)
• High availability of the application
• Fast (or highly compact) data compression and decompression
• Low latency
• Low Environmental impact

(... in all situations ...)

Does performance matters ?

When your website or application is broken, nobody will use it ;)

The slower your website or application loads and displays, the less people will use it

Expiremental results

Some quantitatives results from the industry

• Google – A 500 millisecond delay caused a 20% drop in traffic
• Amazon - Every 100ms of latency cost them 1% in sales.
• Shopzilla – Speeding up their site by 5 seconds increased the conversion rate 7-12%
• Mozilla – Shaving 2.2 seconds off their landing pages increased download conversions by 15.4%
• Netflix – Adopting gzip compression, resulted in a 13-25% speedup and cut their outbound network traffic by 50%.

Is it only for the end-user ?

Do we have a performance issue ?

Most of the time, performance issues are not visible...

• Website/App is up
• No errors
• No alerts
• No remarks from your manager(s)
• No complaints from users

We can suppose the system is working well :(

...But in fact we just don't know

Working on performance
=
Showing what's not visible

How many 3's ?'

Anscombe's quartet

four datasets that have nearly identical simple statistical properties but very different when graphed

Averages are bad !!!

We don't want something to be fast on average, we want it to be fast for most of the users

Most people have more than the average number of legs - Hans Rosling

Use Percentiles !

Why ?

Page loads in more than the average for 30% of the users, even more than 21 secs for 5%

Working on performance

Where ?

You won't see ANY performance issues during unit tests

You won't see ALL performance issues in Development

You can’t optimize everything, you will never have the time to work only on performance

Measure !

Measure !

Measure !

Don't try to guess, never automatically apply tips, best practices... optimizing performance is meaningless if you do not have effective tools for measuring it.

Is this performance issue a real problem for you ?

The art of Micro-benchmarking

Benchmarking is hard, it’s very easy to end up not measuring, what you think you are measuring

common pitfalls  : Cold-start, Dynamic selection of benchmarks, Loop optimizations, Dead-code elimination, Constant foldings, Non-throughput measures, Synchronize iterations, Multi-threaded sharing, Multi-threaded setup/teardown, False-sharing, Asymmetric benchmarks, Inlining

Do not implement your own test rig

Load testing

It's great if your site performs well for a single user..

But how will it do when a storm of users hits?

Every software has characteristics

The throughput is not linear : it depends on the load.

The response time increases proportionally to the user load.

Use the right tools with the appropriate methodoly

In 2017, Why do we still have performance troubles ?

Lack of -production- monitoring

Misunderstood quote

We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil. Yet we should not pass up our opportunities in that critical 3%

Donald Knuth

dependencies

• Third-Parties
• External librairies

Non-scalable design

Not mastering the platform/tools

Lack of ownership

(Unexpected) Big Data

(Unexpected) success

No performance tests

No goals, No objectives, No SLA

Tip : use averges percentiles

No exploration

A Natural evolution of a system

As the software evolves, Performance regression is

expected

Just gave up

Never give up your performance accidentally

Rico Mariani

Building a performance-minded team

• Understand your critical path
• Understand your prod activity & usage
• Do effective testing
• Setup Perf Infrastructure & Automation
• Become metrics-oriented
• Define SLA on what matters for you
• Do regular code reviews
• Educate your team, company
• Celebrate your success !

Your homework

• Can you give me the main bottleneck of your system ?
• Do you have performance/load tests ?
• Can you show me your performance dashboards ?