Please enable JavaScript.

Coggle requires JavaScript to display documents.

Oreilly Conference (Intro (sponsors, tickets: bronze, silver, gold,…

- - - - Title: Practical examples of serverless architecture
        About: Visions of future
        Presenter: :star:Scott Davis (ThoutWorks)
        Serverless emerges from could business. No more concern about deployment, just hack all day. Use cases for serverless:
        
        IoT
        
        Alexa and others
        
        disabled people, robots
        
        Web-development employed for all human senses.
        Apache OpenWhisk for local development.
        Summary: Serverless is important part of web-development revolution for next 10 years, hack all day
        My rate: :star::star: people were disappointed, no code
        Links: Preview
      - Title: Apache Spark and machine learning on microservices
        About: ML delivery pipelines
        Presenter: Stephan Pushkarev (hydrosphere.io)
        How to prepare structured solution for ML data and delivery process. About scripts, technologies and so on.
        Not useful for me yet, sorry.
        
        Links: Preview
      - Title: Research-driven development: improve software you like while staying productive
        About: Solution delivery as a scientific process
        Presenter: Philip Winder (Winder Research and Development)
        Searching for the right solution is very similar to scientific process of checking a hypothesis. It is an ongoing process.
        It is very important to:
        
        setup goals and clarify each word in each statement
        
        ask a lot of WH-questions
        
        define metrics are key, they evolve during single cycle
        
        may require a lot of data
        
        when to stop the experiment? fail smart (not fail-fast)
        
        monitoring of developers empowerment and stakeholders belief in success
        
        if too fast - people do not do their best
        
        if too late - motivation drops
        
        Funny:
        type in google "why software engineer are so" - arrogant
        Summary:
        My rate: :star::star::star::star::star:
        Links: Preview
      - Title: Scaling traditional integration in an event-driven microservices
        About: Real-life story
        Author: Karun Japhet (ThoutWors)
        They built a monolith on an university and come up with performance issues that could not be resolved without architectural change.
        Key benefits of tranformation to event driven:
        
        decoupling
        
        smaller transactions, finished in queue
        
        vendor abstraction (star-trek analogy)
        
        throttler
        
        blue-green swap deployments
        
        Summary: always plan for failure
        My rate: :star::star::star::star:
        Links: Preview
      - Title: Rethink micro-services with stateful streams
        About: Architecture with Apache Kafka
        Presenter: Ben Stopford (Confluent)
        At some point all business outgrows to monolith. Problem is about people and data, not scaling.
        There are some half-truths about traditional approach to micro-services. Two traditional approaches to architecture:
        
        own database
        
        shared database
        
        Both at some point lead to problems. Reasons given.
        Streming Platform is a solution.
        
        lower coupling, instead of calling web-service, listen to events
        
        KSQL to join event streams
        
        new web-service added, but data is already there.
        
        Kafka is designed to handle big Internet workloads and handle disconnectedness.
        Summary: Good architecture is not about <drawing> but about ability to change
        My rate: :star::star::star::star::star:
        Links: Preview
      - Title: The computer science behind modern distributed datastore
        About: Algorithms
        Presenter: Max Neunhoffer (ArrangoDB)
        Modern databases are scallable and distributed and run on much different hardware than old relational databases.
        Key algorithms must be changed:
        Paxos protocol to elect mast and slaves
        Parallel sorting algorithms on new hardware to use L2 cache
        LSM-trees replaces B-trees to support both sorting and bulk inserts
        HLC: clock sync problems
        Distributed ACID transactions problems:
        How to make sure that:
        - all nodes agree that transaction happened? (Atomic) - create consistent snapshot (Consistency) - hide ongoing activities (Isolation) - handle lost nodes (Durability)
        MVCC instead of ACID
        With ACID: Arango (almost), Spanner, Crockroach
        Without: all others i.e. Elastic, Dynamo, Hbase, Mongo, ...
        Summary: Scallability is more important than ACID
        My rate: :star::star::star::star::star:
        Links: Preview
    - - Title: Apache Kafka and the rise of streaming platform :pen:
        About: Streaming platforms is the new future of storage
        Presenter: :silhouette: Neha Narkhede (Confluent)
        High-throughput, low-latency platform for handling real-time data feeds written in Scala. It was successfully implemented in some companies, like Cisco, Uber, NetFlix.
        It is a new value, that can be seen by 3 lences:
        
        Messaging done right
        
        Hadoop done fast
        
        ETL adn Data Integration as a platform
        
        But it gives more than 3 above. It handles more than 100TB data.
        Summary: Streaming Platform is a new type of software and brings revolution similar to Relational Databases
        My rate: :star::star::star::star:
        Links: Preview, Presentation
      - Title: The move toward modularity :pen:
        About: : Transforming monolyth to microservices
        Presenter:silhouette::star: Mark Richards (independent)
        Analogy: going to modularity is like horse-jumping.
        
        Learn to ride - split monolyth to independent modules, but not yet microsevices, just use languange features. No need for organizational change
        
        Learn to ride fast - choose a submodule to split to microservices, but not an admininistrative module, but core business module. Implement some microservices.
        
        Jump - do the organizational change and transform other modules.
        
        Summary::checkered_flag:
        Go to microservices in your company, but no need to transform everything.
        My rate: :star::star::star::star::star:
        Links: Preview, Presentation
      - Title: Architect as a story teller
        About: Architect skills
        Presenter: Natchaniel Schutta (Ntschutta.com)
        Architect is bridge between different departaments, must operate different vocabulary when talking with i.e. technical people and business people or customers. Finding analogies helps a lot. Remembering historical decisions too.
        Summary: ??
        My Rate: :star::star::star::star:
        Links: Preview
  - - - Title: Thinking about platforms
        About: The business value is in platform
        Presenter: James Lewis (ThoutWorks)
        Raise of platforms that provide overall service regarding a problem, like AWS for hosting. The same words are used in context of platforms and microservices, but understood differently and it is hard to communicate.
        Summary: We should consider rather platforms than microservices
        My Rate: :star::star:
        Links: Preview
      - Title: We cannot engineer the future of software without happy developers
        About: All about people.
        Presenter: Yiannis Kanellopoulos (SIG - sponsored by SIG)
        Software engineers should be human. Companies should care about them.
        They need trainings, time for experiments with new tools, etc
        Summary: Good code => less bugs => happy developers => company profits
        My rate: :star::star::star::star:
        Links: Preview
      - Title: The death of Cannot Reproduce
        About: Testability, trackability, monitoring
        Presenter: Michelle Brush (Cerner Corporation)
        In old days she was testing a device with preloaded software so it was easy to reproduce bugs.
        Now we have distributed services and it is very big problem, i.e. 12k bugs on Apache project closed with "cannot reproduce"
        Developer answer for a bug is usually "please provide more details". Testers spend vast of time trying to reproduce, but the ecosystem is changing so it is impossible. Big waste of time for companies.
        Solution:
        
        developers work together with testers
        
        application architecture supports reproducibility of the bugs
        
        They implemented Hoper - simulator of fast clicking on mobile devices, if something goes wrong there are reports that allow for reproducing.
        Summary: Good architecture allows for tracking and reproducing bugs
        My rate: :star::star::star::star::star:
        Links: Preview
      - Title: How to break the rules
        About: Solutions analysis
        Presenter: Dan North (Dan North & Associates)
        There is a book "The Goal" (1984), written before computer revolution, that is still valid. Each solution solves some thing and also settles some new rules that must be followed. It becomes new culture.
        I.e. Continuous Delivery solves problems with delivery but the rule is - we need scripts.
        Other examples like ERP, microservices
        If we adopt new technology we should also prepare our organization for new rules. Otherwise it can prevent us from doing something. We say "oh, there is a lot of work, it is so complex.."
        Summary: Technology can bring benefits only if it diminishes limitations.
        My rate: :star::star::star::star:
        Links: Preview, The Goal - wikipedia), lecture
    - - Title: Scaling and operating ML solutions
        About: ML beyond learning process
        Presenter: Brian Sletten (Bosatsu Consulting)
        Delivery of the first version of an ML solution is relatively quick.
        Adding new features in the system implies changes in the model, need for new data generation, etc.
        There are problems with models isolation, comparing, maintenance of multiple data models and teaching data gathering human feedback.
        Suggestions for data maintenance:
        
        treat as assets with versioning, etc
        
        capture anomalies early
        
        data errors lke code errors
        
        support continuous training
        
        ML is still new. In software we have CI, CD, CD, etc.But we already have something like:
        TensorFlow, opensource from Google. It is an extensive suit not out-of-box solution. It has nice data flow abstraction and visualization.
        GPU are extensively used for learning, CUDA (Compute Unified Device Architecture) by NVidia.
        We need abstraction layers for programming on higher abstration level,like:
        
        OpenCL
        
        WebGPU
        
        ML optimization ideas:
        
        binary file formats
        
        multithreading and queues
        
        3-dimentinal scaling
        
        and so on...
        
        There are OReilly books on topic of ML.
        Summary: A bunch of good practices (?)
        My rate: :star::star::star:
        Links: Preview
      - Title: Serverless architecture: What, why, why not and where next
        About: pros and cons of serverless
        Author: Mike Roberts (Symphonia)
        Serverless is cheap and gives developers change to 'just code'.
        however as more people try we see also problems:
        
        vendor locki-in
        
        bad tooling
        
        latency vary
        
        cannot be statefull
        
        not for long running
        
        resources limitation (RAM)
        
        limited configuration of underlying layers, like OS, etc
        
        Examples how to transform architecture to serverless with API gateway.
        Book provided.
        Summary: Embrace the hybrid serverless architecture
        My rate: :star::star::star::star:
        Links: Preview
      - Title: Software architecture for developers
        About: Good technical leadership
        Presenter: Simon Brown (Independent Consultant)
        Planning problem: need balance between total planning and no planing (Waterfall - Agile)
        Tech-lead needs to be able to explain why solution will work, nt just I will implement everything so it will work (Architecture-aaS).
        Every team needs technical leadership, 2 patterns:
        
        single architect for chaotic team
        
        share the role if team of stars
        
        Tech-lead is typically a good developer, cannot be rank!
        Tech-lead should at least take parts in code reviews.
        Not good situation if architect doesn't code for 20 years.
        Diagrams documentation: need balance between UML (too difficult) and chaotic rectangles. Diagrams should follow common set of abstractions.
        
        Summary: Good architecture enables fast change
        My rate: :star::star::star::star::star:
        Links: Preview
      - Title: practical Kubernetes
        About: Kubernetes architecture
        Presenter: Bryan Moyles (Google)
        Google released Kubernetes as opensource after 15 yesrs of playing with containers.
        It become standard for production deployment for Docker containers. All cloud providers implemented it.
        It is a complete solution. It enabled hardware resource allocation, scalling, rolling deployments, rollbacks, everything that microservice team needs.
        Too long to discuss details. Ii simply need trying.
        Summary: It is the most mature and advanced container orchestration solution
        My rate: :star::star::star:
        Links: Preview
      - Title: Pragmatic event-drive micro-services
        About: micro-services with message brokers
        Presenter: Allard Buijze (AxonIQ)
        Applications tend to evolve to monoliths or ball of mug. Ball of mug has no structure at all.
        Queues allow decoupling.
        Concept behind Axon: command and query responsibility segregation (CQRS)
        Discussed in details.
        Hard to do messaging right. Bounded contexts - clear borders where different models exist.
        Summary: AxonIQ is a balance between very basic message brokers and highly configurable buses (ESB for SOA)
        Links: Preview, Presentation