Excerpt from Andromedia's technical overview of ARIA:
andromedia.com
Intelligent Profiling from ARIA
ARIA takes a radically different approach to the problem, which we call "Intelligent Profiling." In the log analysis systems discussed above, much of the data stored is redundant. In ARIA's embedded object database [ObjectStore], user profiles and web content profiles are immediately updated each time any activity occurs on the site. As a result, that information is immediately accessible for ARIA's Reporter, or for any other time-sensitive application of the information. Databases for Intelligent Profiling are considerably more space-efficient than for log analysis systems (whether the log messages come from server logs, from a server plug-in or from packet-sniffing). On the largest websites, with millions of page-views per day, the ARIA database can be as small as one one-hundredth the size of the relational database required for analyzing log messages. Smaller databases make real-time data retrieval a reality, and database administration costs fall to virtually zero.
[...]
Technical Features of ARIA
ARIA Monitors
Data Captured
The following data is streamed from the ARIA Monitors to the ARIA Recorder:
Client Request Data
Client Post Data
Server Response Data (including message status, such as OK, stopped request, etc.)
Document Meta data
Cookie Data and user profile data
Custom data (e.g. Verisign CZAG encrypted demographics data, user registration information)
Any additional data inserted into server/network data structures
[...]
Efficient data pre-processing
Today's high-volume sites can easily generate up to 400 megabytes of log data per hour. However, much of that information is redundant and/or meaningless.
Taking full advantage of object technology, ARIA creates databases that are orders of magnitude smaller than those created by the nearest competitor. In a log file, full user and content information is stored for every hit to the website. Worse, typical web pages consist of 5-10 hits for each page. That means that a 10-page visit to the website will create a log file with 50-100 descriptions of web content, and 50-100 repetitions of the same user information. The next user that hits those 10 pages will generate an equivalent set of redundant data. ARIA's approach to data storage uses intelligent pre-processing to ensure that each object description (content and user) is recorded only once, with statistical objects updated as appropriate.
Caching
The ARIA database architecture is based on Object Design, Inc.'s ObjectStore database. The data objects are cached before being written to disk in order to streamline the database-writing process. At peak load times, the data can be buffered for later processing during off-peak hours when more system resources become available.
Damien, what do you think about ODIS leveraging the ARIA+ObjectStore bundling to get all the data of Andromedia's customers stored in ObjectStore? I mean ARIA today allows users to stick with their own favorite DB-flavor (SYBS, IFMX, ORCL,...) while storing their web-related meta-data into ARIA's embedded ObjectStore. But it would make more sense to use one single ObjectStore database to store both the raw data and the meta-data in two different subsets!
Hence, ARIA is a perfect tool to increase ObjectStore's pervasiveness into mainstream corporations: first the web stuff, and later everything else! |
