startup technology pitfalls and how to avoid them

39
Startup Technology Pitfalls and How to Avoid them

Upload: dale-whitehead

Post on 17-Dec-2015

217 views

Category:

Documents


0 download

TRANSCRIPT

  • Slide 1
  • Startup Technology Pitfalls and How to Avoid them
  • Slide 2
  • BRYAN SHORT Originally from California Graduated from UCLA 9 years at Microsoft Office Windows Casual Games Xbox Live
  • Slide 3
  • Slide 4
  • BIG BETS
  • Slide 5
  • Architecture Technical Stack Storage
  • Slide 6
  • BIG BETS - Architecture
  • Slide 7
  • BIG BETS - CLOUD
  • Slide 8
  • EC2 Route 53 RDS S3 Glacier CloudFront MORE!!!
  • Slide 9
  • BIG BETS - CLOUD CloudFront Pricing for Dedicated IP Custom SSL is simple. Because of the added cost associated with dedicating IP addresses per SSL certificate, we charge a fixed monthly fee of $600 for each custom SSL certificate you associate with your CloudFront distributions, pro-rated by the hour.
  • Slide 10
  • BIG BETS - CLOUD In Search of New CDN
  • Slide 11
  • BIG BETS - CLOUD Requirements Host CDN powered by S3 SSL custom cert required. Fast and Performant Price Doesnt need to be fastest
  • Slide 12
  • BIG BETS - Technical Stack
  • Slide 13
  • Requirements Able to service http requests Community of active developers Open source friendly Enables engineering excellence Modern
  • Slide 14
  • BIG BETS Mongo? Node / Angular / Express all made total sense. Our data is inherently very relational. What to do?
  • Slide 15
  • BIG BETS Technical Stack How did we do? JavaPHPRubyJavascriptPython 20098573357714513678311210 20102388814910267522751519417 20118192183093827288192166052 2012406197270583478817452320264669 2013 328659255883405824529878210953 2014 220319132449196914280972108289 78, 701 Packages available on NPM!
  • Slide 16
  • BIG BETS Technical Stack How did we do? Not so good things: NPM unreliable Not all modules created equally!
  • Slide 17
  • BIG BETS Technical Stack A Tale of Two Modules Module 1Module 2 Days since last commit6020 External DependenciesNoYes Open Issues17418
  • Slide 18
  • BIG BETS Technical Stack Deployment If adding a single dependency to deployment overcomplicates it; than your deployment process is broken.
  • Slide 19
  • BIG BETS Technical Stack Deployment Like most engineering, its easiest to do the wrong thing.
  • Slide 20
  • BIG BETS Technical Stack Integrated Pipeline
  • Slide 21
  • BIG BETS Technical Stack Test Code Adds incredible value when we you have a product which has logic and behavior to validate. Inside Social is lacking in test code.
  • Slide 22
  • BIG BETS Technical Stack Test Code
  • Slide 23
  • BIG BETS Technical Stack Test Code How are we doing?
  • Slide 24
  • BIG BETS Technical Stack Test Code Corollary! Code that we put onto our clients sites needs to be tested, and ready for prime time production. The worst thing that we can do is break a customers site.
  • Slide 25
  • BIG BETS Technical Stack Test Code Corollary! I am amazed at how many brands, lack any sort of test site / infrastructure and just drop us onto their production site.
  • Slide 26
  • BIG BETS - Storage
  • Slide 27
  • Short term
  • Slide 28
  • BIG BETS - Storage Short term Requirements: Handle lots of writes quickly (implied in memory) Be able to serialize itself, such that state can be restored on another machine SDK
  • Slide 29
  • BIG BETS - Storage Short term How did we do? One month later Aerospike released V3. December 16 th, 2013 Amazon unveils Kinesis service.
  • Slide 30
  • BIG BETS Data Store Data Store
  • Slide 31
  • BIG BETS Data Store How did we do? http://techcrunch.com/2014/03/25/google-launches-bigquery-streaming-for- real-time-big-data-analytics/
  • Slide 32
  • BIG BETS Data Store How did we do? A UW big data study:A UW big data study: BigQuery is very easy to setup and run queries and does not require any manual conguration of clusters; it automatically scales up according the dataset size. But this can be a disadvantage as well, since the user cannot tune the system according to his/her needs. However it has limited SQL language support and does not scale up well on complex queries involving multiple joins and nested subqueries.
  • Slide 33
  • BIG BETS Data Store How did we do?
  • Slide 34
  • BIG BETS Data Store How did we do? What are we missing? Full SQL Data Removal / Archival What works well? Pricing NoSQL
  • Slide 35
  • BIG BETS - Architecture
  • Slide 36
  • How are we doing? Were able to accomplish our mediate goals! Great Success!
  • Slide 37
  • BIG BETS - Architecture How are we doing? Can we solve our long term problems? Real time? Complex system that evaluates what social marketing tactics, and learns from them?
  • Slide 38
  • The road ahead Once we have something stable, that people can utilize it is time to revisit some bets. Inside Social will be reexamining short term data store + data store options.
  • Slide 39
  • Thank You ? Any Questions