Nuts and Bolts of High Quality Internet Applications Talk

Site Home : Teaching : Short Talks : One Element

Introduction

• old world: desktop applications, one user at a time at the mouse and keyboard
• Web world: many users simultaneously updating a common database
• modern world: Web world + servers talking to each other
• Today: perspective for engineering design meetings, inspiration for the future

how an Internet server communicates with the rest of the world: HTML, VoiceXML, SOAP, WSDL

• what's interesting about HTML: markup language instead of programming language; you specify the application and the browser renders the application. Result: lot of innovation in computer apps accessible from a desktop computer.
• VoiceXML: same deal but for voice
• mobile: lots of protocols, latest is XHTML Mobile Profile
• SOAP, WSDL: server-to-server communication

how an Internet server handles simultaneous updates from 1000 users at the same time

• the relational database management system (RDBMS) and the Structured Query Language (SQL) -- versus spreadsheet program (guy in closet analogy)
• declarative versus imperative
• use of imperative languages has kept programmer productivity low; only way to improve productivity (output per dollar of labor) is moving jobs to India
• you can do this with all computer problems because Haskell is Turing-complete

effective page design: screen space, time, words, color, standard controls

effective interaction design: navigation, user testing

two-tiered versus three-tiered architectures

• reliability: best to have one simple physical computer; when was the last time you saw a broken desktop PC?
• challenge of scaling: hard to predict load on a public site, may not be able to serve whole world from a one-board PC
• draw lots of computers on board (99 percent reliability implies 10 machines have 90 percent uptime, 100 machines only 37 percent; Google has more than 10,000 computers)
• draw layers on blackboard:
  • transport-layer encryption (SSL if the site has secure HTTPS pages)
  • HTTP service
  • presentation layer (page composition; script execution)
  • abstraction provision (sometimes called "business logic"; any layer of code on top of the raw database where each procedure is used by more than one page)
  • persistence
• start with persistence layer: better to have lots of CPUs in one box or many separate CPUs? Depends on how many updates database is getting (communication through shared memory very fast).
• abstraction layer, where people often add a tier. Banks used to add application server programs because a transaction touched three databases (checking account, credit card, customer service database). Orbitz has a layer here because searching for flights computationally expensive.
• HTTP service: can be CPU-intensive with things like Active Server Pages; where to put all the CPUs?
• reasonable compromise for most sites: multi-CPU database server, multiple Web servers, load balancing router; can upgrade the operating system on the Web servers without touching the database. Expect no better than 10 pages per second per CPU.

the most productive software development styles

• the real design work in an Internet application is in SQL and the page flow (interaction design); imperative programs are just glue between the database and the Web
• the Microsoft Active Server Page style of development is very effective because all of the code for a page is in one or two files; easy to make changes
• systems programming languages have very powerful type systems (C++, Lisp, Java) but these aren't useful when source of persistence is RDBMS (limited to string, number, and date)
• if source of persistence is an object database a language like Java would make sense

can you start with an open-source toolkit?

A little inspiration

• old days: programmer could only have impact as industrial worker (50 people to get a CD-ROM in a package to the consumer)
• good news about modern times: one programmer can distribute via Internet
• bad news about modern times: only one software company (Microsoft monopoly)
• you can't be smarter, harder working, or slicker at marketing than MSFT
• you can pick a problem that they haven't thought about and get a solution into user hands ASAP
• standard product development/release cycle: 2 years, users to marketing to product managers to programmers back to users
• what we did for ACS: shortcut by release from a running system