Todd McKinnon, (ex) SVP for Software Development, Salesforce.com (now http://www.saasure.com/) gave a talk at Palo Alto Research Center. Searching the net my sense is Todd said more at PARC than at other forums. PARC is about sharing research conclusions, salient material is conveyed to get further understanding via Q&A from peers.
It's taken me about a year to grip the significance of the Salesforce.com architecture and the possible impact on social persistent shared business apps. Salesforce solved the problem of building CRM application Entry Forms from collections of private customer data in multi-tennant horizontal partioned web database.
Salesforce architected the fundamental architecture around the 2000 from lesson learned at CommerceOne long before the basis for scaling social web apps commonly became architecture patterns from search: Bigtable / MapReduce / Hadoop / Sharding. Thus Salesforce uses Orcale databases in a very unconventional pattern. Database on a Database and custom indexes (not Oracle indexs).
My conjecture: It may be possible to build multi-tenant business apps without strong schemas for certain shared business data by mining for key value pairs from transactional data written via entry forms in linear time and later mined for associated collections via a search interface.
Hybrid systems could emerge. ERP private business data in tradtional relational storage. (Write once update many times). And socially purposed shared data stored in a Hadoop pattern. (Write once / read many times).
Applications would need to be natural cases for sharing data such as Lot Serial Traceabilty across the Food Supply Chain.
The FDA and Public (why not, the public) could trace peanuts to farms, peanuts to contract mfgs, food products containing peanuts to distribution centers all the way to retail stores and your food snack. The FDA could probably improve traceabilty by sponsoring University/ACM research in this area.
Driving a bus between the details in this conjecture is easy, what I'm looking to find is grounds for innovation from new data patterns.
Compare the Salesforce metadata table schema to a map in HBase/Bigtable. 
Force.com Multi-tenancy White Paper - via Todd McKinnon
Understanding HBase and BigTable - Jim R. Wilson
Todd's talk at PARC, excellent write up - Paul O'Rorke
No comments:
Post a Comment