Scalability Lessons

Danny MacAskill's Imaginate

Beyond 50k 85k requests/second

Your Host Tonight

Image source: my 5 yo daughter

Software engineer at MediaSmart Mobile

Freelance scalability consultant

  @pinchito

What We Do

Serve mobile ads

Performance campaigns

85K+ requests / second

10M+ impressions / day

40+ servers

20+ countries

Guilty!

Image source: Stupid Zombies 2

We help pay for your entertainment

From the particular to the abstract

• Scalability with Node.js
  



• Effective Engineering in the Cloud



• Zen of Scalability

Scalability with

What Is Scalability

Serve an increasing numbers of requests

• with increasing resources


• without degradation


• cost-effectively!

Scale Up vs Scale Out

Learn how to use both, and when to use each

vs

Image source: Rama

Image source: ZDNet

Is Node.js Fast?

Define ‘fast’

Source: GodTIC (Spanish)

So Why Node.js?

Asynchronous, event-driven, non-blocking

Focus on concurrency, not performance

Node behaves very linearly:

Double the load, double the resources

Smooth learning curve

Few surprises

Plus, it's cool...

100K+ Modules to Choose From

Source: modulecounts.com

Far above all other package managers

Linearity Rules!

Effective Engineering

Image source: Beao

in the Cloud

Load Testing

Gather data instead of opinions

Two kind of tests:

• real (production)
• and fake (load tests)

Measure requests per second

and convert to visits per day

... with a safety factor

Test to See Where It Breaks

Image source: MythBusters

Do not be afraid to work on production if needed

Keep Up with Business

Keep business happy

And we will all be happy!

Unpaid Technical Debts Accumulate

Image source: BuffaloWorks Photography

Developments go slower

Your code runs slower

Managing Technical Debt

Business Goes First

Do not scale too soon

... and definitely not too late!

Keep lists of what you owe

and repay it as you go along

Don't let performance issues accumulate

Deployment

DYOD: do your own deployments

Deploy as often as possible!

Monitor after deploying!

Monitor everything!

Send alerts

Avoid False Alarms

Image source: DIYLOL

Send only necessary alerts

Remote Caching

Humanity's best invention

for scalability

Cache everything everywhere!

Locally and remotely

Memcached + node-memcached

  Redis + redis driver

Messaging

Humanity's second best invention

for scalability

Ideal for async processing

Remove work from online processing

and aggregate it

RabbitMQ + node-amqp

memcached + RabbitMQ Architecture

Zen of Scalability

Image source: CloudBees Blog

there is no Perfect Architecture

It's always about trade-offs

Keep your architecture fluid

Locks are Evil

Always

No exceptions

Search for lock-free solutions

Smash those Bottlenecks

Scalability is about removing bottlenecks

Possible Bottlenecks

• CPU (obviously)
• memory
• disks
• network interfaces
• system buses
• connections
• messaging
• latency
• simultaneous requests
• frontends
• backends
• <insert your own here>: throttle down traffic

Saturation

Is the sign of a bottleneck

Beware the Thundering Herd

... and beware the self-sustaining stampede!

You Found the Cause...

Now try to find the real cause

Do root cause analysis immediately — and fully

One More Thing: Be Humble

The world doesn't care about your feelings

You cannot win an argument with reality

Unverified opinions are useless

(specially your own)

Accept facts and collect data

Thanks!

@pinchito

http://slid.es/alexfernandez/scalability-lessons-beyond-85krps