Notes from AWS User Group Dublin

Update 10th March 2018: Alan Kiernan of Cation Consulting reached out with some additional notes on his talk, added below. I've also added an amendment to the first overview section for clarity purposes.

 

Interested in reaching out to discuss more - the Twitter thread is here or else you can comment below.

IMAGE.JPG

Overview

As part of my usual inquisitiveness, I've been trying to expand outside my usual circles in the telecoms world to see what is happening related tech sections. In the past, this has led to attending some Virtual and Augmented Reality events (2017 notes and 2018 notes) , as well as expanding on some historical events.

The main reason is it's too easy to stay in your comfort zone, only hearing about what is going on within your circle with the only information you hear elsewhere coming from books and websites. The other is because, as I've learned from my sport of climbing, there is usually other sports going things a lot better than you and instead of re-inventing the wheel and starting from the absolute bottom, why not catch up to where the current best-of-breed is by listening to what they're doing?

Therefore, March 6th 2018, I ended up in the Button Factory in Templebar being handed a free beer and food to hear several very interesting talks from a range of speakers. It probably also helped that AWS Builders Day was on the same day so it meant for a strong collection of AWS experts in the crowd also!

 

My general takeaway: infrastructure is now a commodity and your mindset about capacity and load in the modern web era should be reset. Every service has the potential for a global audience and as highlighted the Collison twins and Stripe, it now only take 25 months to get to 100 million users - the idea of 1 million users isn't an achievement now for a hyperscale platform company. Along with that, it was evident from the Alexa demos that even adding voice 'Skills' is becoming relatively trivial, and while it's early days, people are already making (likely decent) money off the platform. The other takeaway and putting my telecoms hat on: connectivity isn't ever discussed at any of the tech events I've bee to - it's just assumed as a given that the connectivity is there and it's fast and responsive. [Addition/clarification, 10th March 2018] That should also be seen as a positive and an opportunity - that means every single device needs a connection (some directly to the Internet, and some aggregated on a home or corporate connection. Lots of opportunity here!

[update 10th March: another interesting consideration, especially in light of the current active political climate around child safety online. If all these current kids are growing up speaking to Alexa and having a device in the house that is listening for a keyword, what would they say about that and child safety/privacy? (personally, I think privacy means something different in the Internet-era).

 

 

Notes on the AWS User Group

AWS Dev Group is now three years old. 3,000 members. 

The objective of the group: "focus is on making something people want". 

 

Live streaming the event for first time (using Twitch of course, Amazon's streaming system).

Link to the videos from the night can be found here: https://www.twitch.tv/awsdevs/videos/all

(I've also tried to link to the correct videos below)

 

Speaker 1 - AlertLogic

Recording can be found here.

Observation from the speaker: Innovation with AWS in almost impossible to keep up with. 

Can treat the cloud as almost Infinite scalability. 

Prevalence of web apps. 

 

Cloud Security company: need different approach to cloud. 

 

Principles of security don’t change but the approach does. Need to understand shared responsibility of cloud security. 

 

205 days on average before detection of compromise. Two-thirds of companies find out from a third party they’ve been compromised. 

The slide below I loved: in blue is what is Amazon's/AWS's responsibility, in Orange is the AlertLogic, and white is the customer responsibility.

IMAGE.JPG

Speaker 2 - Rick Harvey, Alexa evangelist

Recording can be found here.

Time to reach 100 million users. 2003-2008 took 72 month, 2009 on-wards now only takes 25 months so infrastructure is critical. Cloud makes much easier..  

Serverless (lambda) billed per millisecond. No servers to provision. 

AWS Fargate - run containers without managing servers. Pay per second. 

 

Speaker 3 - Ronan Gilfoyle, Solution Architect. 

Recording can be found here.

Ronan appears to do an update at every event on the newest services/products released in the past month. Observed that it's getting harder and harder to keep up such is the pace of change.

 

New Services/changes

Trusted Advisor is now free to check security of your services. 

AWS WAF 3 new managed role products. Easily deployed using AWS Application Load Balancer or Amazon Cloudfront. It’s free if you don’t enable rules. 

Reflection attacks. DDoS attack now only needs a dial up connection. 

GDPR - DynamoDB supports encryption at rest. 

Backup - virtual tape library. Support Dell EMC Networker 9.0

 

Speaker 4 - Alan Kiernan - Cation Consulting

IMAGE.JPG

Gave a demo of Alexa from the DevOps world. 

Concept to demo, before the first meeting. Minimum viable product - no infrastructure cost! Pivot without too much sacrifice. 

 

Ryanair Rooms - John Hurley, CTO: "get it going, fast, iterate”.  Can't recommend this enough.

The best of the good stuff.  Lean, agile, start-up mentality. 

 

Discussed the idea of an Alexa Devops chatbot: can Alexa be part of DevOps team? Not just yet.

A very entertaining demo of the sentiment analysis platform (AWS Comprehend) on AWS: by doing a scan of tweets about Donald Trump.

Directly from Alan:

The presentation/demo itself was of Alexa being used to fire off and build AWS Infrastructure in realtime. That infrastructure consisted of the required components, software and integration to capture twitter in realtime, feed to comprehend, elastic search etc etc. So, before the demo started, that infrastructure didn't exist. The key point here is that as part of an Alexa skill, on-demand infrastructure was be "spun up" as required to service the functional needs to the user.

Built the product the night before, and demoed it live on stage: 'Alexa, start social analysis', 'Alexa, start social analysis of Donald Trump', and response from Alexa of 67% negative, 33% positive and all the main keywords. [update 10th March: Alan contacted me to observe I'd - somehow' mixed up the two results. 67% would be unheard of for President Trump!]

 

Speaker 5 - Tito

Event booking system: €300 million sold since creation in 2011, €250k monthly revenue, 400 daily events, 2 million tickets issued.

Gave a talk on how Tito was created to how it has evolved to where it is going.

 

Part 1 - the beginning

The founder built an API on a flight. First customer was Horace Deidu (incredibly smart analyst) for his first conference (big name!).

Largely bootstrapped from 2011 until 2015 - staff also doing consultancy on the side. End of 2015, three people full time and now four.

Currently, 87% customer satisfaction.

Now eight staff (four until 2015), 5 male and 3 female (all developers).

Split between engineering and design, and marketing.

Not hypergrowth, but very steady. [Fascinating that €300 million in revenue over 7 years isn't hypergrowth anymore]

 

 

Part 2 - Philosophy

Balance: bit of flexibility now due to the revenue and growth, but still principles-based.

Always asking: Why are you doing what you're doing?

Focus of Tito: great experience for customers. Product-focused.

 

Diversity and Inclusion - started as four white men. Basically the norm in Dublin, now 3 of 4 new hires are women.

Integrity and respect

Design-led, empowerment, happiness is greater than perfection.

 

Part 3 - the stack

Open-source: ruby on rails, etc.

Automation by using AWS. allowed to far exceed what one person can do.

Testing stack on AWS: immediate feedback. Always through CI.

Load testing (important for a ticket engine!). Web Summit: 25,000 in five minutes.

Long betas

outsource as much as possible.

Tito stack: two HAPROXY (better than ELB as up faster), application (on two app servers, multi-AZ), memcached (job server is sidekicked - commercial version, both are multi-AZ),  DB (Aurora - triple/multi-AZ), OpsWorks intervenes if a server goes and and auto-brings up a server if it goes down.

1.5 hours of downtime in 2017.

 

Heroku on AWS Dublin also. SSO, Billing, QS, Passbook, dashboard, staging - currently splitting out into microservices.

CI on Travis

Dev Stack is on Github.

 

Playing recently with vue.js - it's the future.

 

 

Tito - 1st version just an API, redirect call to Paypal. Future: API forward that is one-for-one with what the UI does.

 

Scaling issue that needs to be considered (flash sales are unique to event handlers). 30 million tickets in an hour for Robbie Williams would have killed Tito (but company who did provide were fine).

 

Documentation helps everyone become better developers. Working to 'push the boat out' on documentation.

 

Questions:

Most difficult part to get up and running? had someone who was really good at UI design. UI design - you need to be extremely patient and careful (should it be 500ms or 100, etc.)

Finding technical staff/designers? the last two jobs that were posted, only two people applied.

 

 

 

Speaker 6 - Intercom - Rich

Engineering Philsophy - run less software.

This talk is about what this means.

 

Went to Patrick Sutton - a speaking coach. How do you want people to feel?

We're at war in technology - but no one realises we're in it.  

 

  • Innovate startups (includes Intercom, naive, dreamy)

  • Copycats (copying people like Intercom, and never been easier to beat you at your own game - money is cheap, VCs are going crazy, if you can point at someone else doing something succcessful you'll get funding, AWS is making it easy to build infrastructure so copying is easy, RubyOnRails/Ember makes it easy to build a working version of a product in weeks or days, basic marketing and discovery is super easy - app store, Play store, )

  • the four horsemen: Apple, Amazon, Google, Facebook. They take a cut of every single battle. they are also competing in war for talent (and a lot of items to disposal: job security, money, etc.) have more credibility, time, money than you (and likely more talented) so if they step into your turf, you have a real challenge.

  • army of talent: demand outstrips supply three to one. speaker is up here today to try and win you over.

 

Example given of HipChat and Slack. Hipchat were first to market, Slack have successfully out-executed them (and Asana).

Instagram and Snapchat another example. The revenue generation engine, not the UI.

AWS against Blue Apron. Amazon just registering a domain name hit the stock of Blue Apron.

 

Worth remembering right now about the current situation

  • Money is cheap

  • Basic execution is easy

  • Talent is scarce

  • Threat from one of The Four (Amazon, Apple, Facebook, Google)

 

Neo in the Matrix: defeats by being able to wield time. "Time well spent is when our top talent is focused and productive solving only our most important and differentiating challenges".

The Three Pillars of Run Less Software:

  • Save Time: choose standard technologies

  • Save time: outsource undifferentiated technologies

  • Spend time: create enduring competitive advantage

Standard technologies: could choose any weapon in older days, but choose one or two weapons to be become excellent at. Generals would then decide what their whole army would use.

Choose boring technology: solve problems by constraining yourself, mostly but not exclusively to solving them with a small opinionated, company-specific set of standard technologies.that over time you become an expert in.This will serve you well in the long run

Total cost = (sum of operations cost) - (sum of velocity benefits)

 

Intercom Core Technologies - Google Doc (owned and maintained by CTO), 10 pages long, continuously updated and curated by multiple people (shared with beginner employees also), color coded:

  • Green = default choice

  • blue = supported but know why you're not using the default, a review is more likely

  • orange = non-standard choice for specific use cases, more review needed

  • red = deprecated / discouraged. lots of convincing needed (to use again)

 

E.g. for server-side execution environments

  • green = main rails.app

  • orange = secondary ruby app

  • orange = Go (performance sensitive services)

  • orange = python (projects consuming sci-py/ML)

 

Intercom - a bunch of their 'standard' technologies are AWS.

 

Outsource undifferentiated technologies

"There is surely nothing quite so useless as doing with great efficiency what should not be done at all". Peter Drucker.

"There is a lot of undifferentiated heavy lifting that stands between your idea and that success. 70% of your time, energ, and dollars go into the undifferentiated heavy lifting and only 30% of your energy, time and dollars gets to go into the core kernal of your idea. I think what people are excited about is that they're going to get a chance they see a future where they may be able to invert those two." Jeff Bezos.

DO outsource your undifferentiated heavy lifting.

DO NOT outsource your Core Value Proposition.

Things to consider when outsourcing: security, reliability, flexibility, cost.

User data - very sensitive! Generally won't let this outside of their AWS systems.

 

How they decide for security reasons to outsource - use the ranking below (Tier 1, 2 allowed see data).

  • Tier 1: AWS

  • Tier 2: best-in-class public or late-stage companies (e.g. Stripe, Sparkpost, Datadog)

  • Tier 3: mid-stage start-ups (e.g. Greenhouse, KeenIO, VividCortex)

  • Tier 4: young start-ups (e.g. Honeycomb, Foxpass, etc.)

 

                        undifferentiated heavy lifting                            core ideas

The industry            70%                                                            30%

Intercom                 40%                                                            60%

 

 

The things you own, end up owning you - Tyler Denton (Fight Club).

 

Intercom's core:

All UI - javascript (owned)

messaging - Intercom Nexus, Intercom Messenger

Search - currently their own baremetal ElasticSearch (considering moving to AWS's hosted solution)

 

 

 

Theory in Action:

Scaling our User Storage System:

1.2 billion monthly active end users

Aurora and Dynamo both are on core technology list. Re-did their user service on Aurora but not natively scaleable. Dynamo didn't do multi-part primary key

Break problems down until they could be solved with standard technologies (mix of Aurora and Dynamo). Build protoytypes to eliminate risks.

90% cheaper (dollar value) and freed up 3 engineers from being called during night to deal with MongoDB issues.

 

What's the prize if you Run Less Software?

Shipping - you can ship a lot of stuff! Great feedback loop, and none of it will involve support calls during the night.

One of biggest SaaS companies of this generation.

 

Intercom Foundation

Intercom. Diversity, diversity, diversity!!!!!!!

 

 

 

 

Speaker 7 - Max - Lead Evangelist for Alexa

Video here.

Build with Alexa: 

Alexa for Business released. One fundamental change is skills were globally available - now can make organisation-specific skills.

Generation of kids who are growing up talking to their house. it's going to be big for the next generation!

 

Alexa Framework: 

check your conversation between human beings otherwise it'll make no sense!

Full explanation at 1:08:00 of the Framework.

 

7 golden tips to develop Skills

  1. Design for the ear (i.e. practise your conversations between real humans beforehand)

  2. use the skills console

  3. good training data

  4. leverage states

  5. persist, persist, persist!

  6. be mindful of cognitive processing

  7. listen to how people talk

 

 

Business idea: Alexa skill to record notes on a customer or client while in a session (link to a file/note/calendar-event/etc.).

New Alexa Skills Kit developer console.

Think multi-modal, screen provides rich experience at times also.

There is a Smart Home skills API

 

AWS Device SDK - now allows you to choose your own wake word engine

 

Shows on-screen 10 different development kits (different mic setups, arrays, etc. etc. - one end is the 7-mic circular array, to other of a single mic)

 

 

Show me the money - how to make money from Alexa

  • Today - Alexa rewards and credits to stimulate the community

  • Late in 2018: subscription model (currently developer preview)

  • Alexa in-skill purchasing

  • Amazon Pay for Alexa Skills

 

Alexa development fund: $100million on offer. Looking for builders around the world.

 

Success Stories:

  • Nick Schwab

  • 2 year old company. published 50+ skills, 2 million monthly unique users.

  • 2 Petabyte of content

 

  • Opearlo

  • Unilever, Diageo, Coty, Vitality, Panda Rescue, Find my Phone, Easy Meditation, Easy Yoga

 

  • Labworks.io

  • On this day app

  • Would you rather? app