[Scrappy notes from a presentation earlier this week by Jeff Barr to the Open University on Amazon webservices]
(Amazon now has 57 million active customers worldwide.)
AWS started to become reaility 2002.
Amazon is a technology consumer (25 TB of business/historical data, 21 fufillment centres worldwide) as well as a technology provider - to Amazon Associates (1, 080, 000 active sellers) as well as s/w developers (160, 000 signed up for ws).
Programmable website
- support for industry standards
- provide remote access to data and functionality via apis
- decouple data and presentation
- means to leverage technology investment
AWS -
- building block web services (storage, queuing)
- ecommerce capability
- new business models
- customer created content
8 public web services
- Mechanical Turk
- AWIS - alexa web info service - crawl every 2 month, data acquired via Alexa toolbar (300TB of data), 3 types of crawl data: current, historic, new crawl
- AHP - amazon historical pricing - e.g. mine historical data to idenitfy seasonal trends
- SQS - simple queue service - FIFO
- Alexa top sites - top sites by alexa traffic rank
- AWSP - alexa web search platform - roll your own search engine
- ECS - ecommerce service (includes similarity data (?) - people who bought this also bought that)
- S3 - simple storage system
[No mention of AWSZone Scratchpads?]
Each web service seen as a separate business; developers are an important new customer segment for Amazon; some services for free (because business comes back to Amazon); some services are paid for (e.g. S3);
Business model associated with each web service
3 business models
1) free access, 1 call per second, traffic back to amazon; e.g. ECS; data in exchange for sales; developers can also earn money e.g. through amazon associate program; alse need business model for developer
2) monthly fee - e.g. historical pricing service
3) usage/resource based charging model e.g. Amazon S3, alexa services, SQS, Mechanical Turk
- can pay for this using personal amazon account.
ECS
Now in 4th major release
Complete access to amazon product catalogue
Reviews
Shopping cart - checkout on amazon property, tho' cart can be filled on any site
wishlists
e.g. click sharing [I like that:-)] (innovative business model) as seen at http://www.associate-o-matic.com/ - Amazon store builder in which developer takes commission from 1 in 10 transactions
Specialised sites tend to be vertical in nature, more focussed e.g. in terms of market segment.
S3
Pay as you go/pay only for what you use
15 cents per gigabyte month
20 cents per G to access data
Private and public storage
scalable, reliable, cost-effective, simple for developer
800M objects stored to date
Throw any amount of data at it, and that's fine :-)
Web scaleable, automatic replication
4 9s reliable
Same storage facility is used for Amazon product catalogue
Amazon don't peek inside stored data - user encryption okay
Anonymous storage not possible - user must register to be able to use service (so they can be billed...)
Linear cost scaling (cf. having to over-invest in your own servers against predicted grwoth of your web business)
Data - 1 byte to 5GB
REST or soap api; unique URL for each object in s3
Amazon Mechanical Turk
Tasks that are "easy for a human, tough for a computer"
"people on demand"
Make the most of latent labour workforce.
Looks like a machine, but actually relies on human information processing inside ("human inside", "person inside") - artificial artificial intelligence; human in the loop; HITs - Human intelligence tasks
Developer can put a qualification step in the workflow e.g. to test the worker is qualified to performn the task.
May get the same job done by several people, then if 2 from 3, 3 from 5 etc give the same answer you have some sort of reliability guarantee over quality of the work
Reputation is tracked - number of HITs applied for and number of hits accepted on worker side; number of hits completed and number of HITs accepted on job allocation side
Qn: 1-click as paypal alternative? not as far as aware
REST vs SOAP - as of a year ago, about 80% REST , 20% SOAP
I meant to ask: "Amazon ECS in SL?" but didn't get a chance. However, it seems someone has just hacked the search part...
Extensions - e.g. book burro on Jeff's Amazon catalogue pages...
Okay for browser extensions like book burro if Amazon stands a chance of getting some traffic back as a result.
[??is this the same for using webservices? e.g book details completion on OU library catalogue? - maybe put a link to 'buy this book on Amazon.]
All in all, an interesting overview (particulalry some of the numbers and the three business models for ws decomposed) and good to see someone I read via the Amazon web services blog in the flesh.
Posted by ajh59 at July 27, 2006 10:59 AM