Download to read offline
Uploaded byBenjamin Bengfort
Digitocracy without Borders: the unifying and destabilizing effects of software internationalization
The document examines the evolution of software, emphasizing the shift from traditional products to software as a service, highlighting the importance of internationalization and user growth metrics. It addresses the impacts of open source, machine translation, and the future of global consistency in distributed systems. The analysis also touches on the challenges and advantages of geographic distribution within software infrastructures and the legalities surrounding data management.
More Related Content
Similar to Digitocracy without Borders: the unifying and destabilizing effects of software internationalization
Digitocracy without Borders: the unifying and destabilizing effects of software internationalization
- 1. Digitocracy without Borders TheUnifying and Destabilizing Effects of Software Internationalization
- 2. Software Economics Software isno longer treated as a traditional, consumable product that can be shipped to stores and customers in boxes, purchased routinely and frequently. Instead, software is a service and is generally expected to be free or mostly free.
- 3. Software Economics Software monetization strategiestherefore rely on “hockey stick” user growth curves that rely on network effects. With enough user density, ecosystems form that support digital products.
- 4. Software Economics In acompetitive marketplace, apps must target international audiences: there is no such thing as “farm to table” software. Most software is built with localization best practices from the start.
- 5. Traditional political and economicinstitutions are constrained by geography and ideology: they cannot regulate panopolies. We have become more connected than ever before and more people have the opportunity to have a better life and benefit from technology.
- 6. What have beenthe effects of software internationalization? The Past Open Source The Present Machine Translation The Future Global Consistency
- 7. The Past: OpenSource
- 8.
- 9. Open Source Extremely complex,very important, and evolving rapidly. - Culture and activism - A mechanism to build good software - An economic model - A commercial model - Technology-related phenomenon
- 10. A Brief Historyof Open Source Hackers and the MIT AI Lab “Sharing recipes is as old as cooking” Free Software Movement “Free as in speech, not as in beer” Open Source “Given enough eyeballs, all bugs are shallow” Affero GPL “Closing the SaaS loophole”
- 11. GNU Emacs Small numberof developers hand curating every line of code and feature. The Cathedral and the Bazaar Linux Kernel Source code developed in public over the Internet with many contributors.
- 12. Yellowbrick NumFOCUS and DistrictData Labs affiliated PyData project for machine learning visualization. 15 releases, 78 contributors, 2125 stars, 357 forks. Libraries.io SourceRank: 14 3 languages translated docs https://www.scikit-yb.org/en/latest/ https://github.com/DistrictDataLabs/yellowbrick
- 13. Yellowbrick Documentation UniqueUsers GSoC PyData
- 14. 2009 Open SourceIndex: Georgia Tech and Red Hat http://jsk-sde.blogspot.com/2009/04/open-source-activity-map.html https://www.redhat.com/en/about/press-releases/open-source-index
- 15. Of companies runon open source; less than 3% don’t use OSS in any way. https://www.zdnet.com/article/its-an-open-source-world-78-percent-of-companies-run-open-source-software/ 78%
- 16. Critical Infrastructure runson Open Source In 2012 the heartbleed bug was introduced into OpenSSL, which was publically disclosed in 2014. Security flaw creates a hole that affects 17.5% of servers and 68% of Internet traffic. OpenSSL is maintained by 1 full-time, underpaid maintainer.
- 17. The Present: MachineTranslation
- 18.
- 19. State of theArt Attention Based RNN (LSTM) Encoder/RNN (LSTM) Decoder network with source and target embeddings. Deep Neural Networks.
- 20. Artificial Neural Networks Arbitraryset of inputs → arbitrary set of outputs Hidden layer of weight functions (neurons) that map a lot of other functions (a series of simple weighted functions)
- 21. Deep Learning Neural architecturesthat embed increasingly complex learning layers to optimize for computation or error minimization. Essentially, brute force function approximation.
- 23. Bringing neural networks tomobile devices Method for modeling neural networks’ power consumption could help make the systems portable. Meanwhile, more smartphones in Africa than ever before. http://news.mit.edu/2017/bringing-neural-networks-cellphones-0718 https://www.gsma.com/r/mobileeconomy/sub-saharan-africa/
- 24. NLP/AI toolset hasfailed Facebook in addressing hate speech, which may have unintentionally made the social media application a framework for genocide. In Burmese ကုလား (Kalar) commonly used as a slur against Muslims ကုလားပဲ kalar-pe (chickpeas) ကုလားထိုင် kalar-tain (armchair) ကုလားတည် kalar-de (pickle). Zawgyi vs Unicode encoding Facebook in Myanmar
- 26. The Future: GlobalConsistency
- 27. Context: A SingleServer Let’s suppose a simple data storage system where clients can: - GET(key) → value - PUT(key, value) - DEL(key)
- 28. Context: A SingleServer The order of the operations determines the responses the system gives. In a single server system, the server can determine any order it wants. It is always consistent. 1 2 3 4 5 6
- 29. Consistency The system respondsto requests predictably
- 30. Context: A DistributedSystem A system is distributed when it contains more than one server that must communicate. If one server in the system fails, it doesn’t necessarily become unavailable because other servers can answer requests. If data is replicated we can also avoid data loss.
- 31. Inconsistency When multiple serversparticipate in the system, they need to communicate in order to ensure they remain in the same state. Communication takes time (latency) and the more servers are in the system, the more time it takes to synchronize. PUT(x, 42) ok GET(x) not found
- 32. Concurrency Because of delaysin communication, it is possible for two clients to concurrently perform two operations - from the system’s perspective, they happen at the same time. The order of these operations determines the response. PUT(x, 42) PUT(x, 27) GET(x) → ?
- 33. Distributed Consensus Fault Tolerant DecisionMaking for a Network of Replicas. ?
- 34. Geo Replication Distributed Systemsof Servers that Span Continents State of the Art: Spanner Academic Research: Alia/Calvin
- 35. Geographic Distribution Things getworse when you try to put servers in different data centers: - Latency increases because of physical limits - Network partitions can cut off groups of servers
- 36. Geographic Distribution There arebig benefits to geographic distribution, however: - Resilient even in catastrophic failure of a data center. - Servers respond more quickly to colocated clients.
- 37. Global Clouds Cloud serviceproviders are expanding their network of data centers across the globe.
- 38. Legalities of GlobalSystems Systems are regulated in the countries where the data resides (where the servers are). Cloud coverages means its easy for service providers to quickly switch the location of their service.
- 40. Project Loon andFacebook Athena
- 41.