IQ is what is called a "column store" database. Instead of storing a record at a time and stacking records into pages and pages into page strings, it stores single column values into pages and pages into page strings.
When data is loaded, each column is treated separately. IQ will light up a thread for each column of the table being loaded. Each thread handles just one column.
In the process of loading a column, the data is compressed and indexed. How this is done depends on the data type, cardinality of the domain, and the index type(s) explicitly defined for that column. For instance, very load cardinality fields such as GENDER (Male/Female) will be defined with a "Low Fast" index. This creates a bit map rather than storing the actual character string and then applies a compression algorithm to the bit map. As a result, instead of taking an average of 5 bytes/40 bits (plus overhead) per column value for GENDOR, it takes less than 1 bit given the compression and you get an index on the column for free.
IQ pages are not really fixed size, either. Pages start out at some predetermined size set at database creation time, typically 512KB. The page holds the raw data for the column and, of course, the number of column entries it contains depends on the size of the raw data. The page as it is stored on disk is the size after compression and indexing. In this way, the actual I/O bandwidth and disk required can be smaller than the raw data size and is always much smaller (usually a quarter to an eight) than the overall space required for a traditional RDBMS after loading and indexing.
Row reconstruction is done by using the ROWID. Column values for a particular row can be constituted into a traditional row because the ROWIDs are tracked for each page.
The speed of IQ comes in part because you need only search the data (columns) that appear in the WHERE clause of a query, in part because every column is (or can be) indexed, and in part because the data is spread out over a lot less disk space.
The outline above is a greatly simplified view of how IQ function internally but it gives you a rough idea of what is going on. For a more detailed look, you can see my article, "An Introduction to Sybase IQ" in one of the ISUG Tech Journals in 2007.
BTW, the nominal cost of membership not only gets you a subscription to the Tech Journal, it also gets you access to a lot of Sybase software for free or at greatly reduced prices. For instance, you can get a fully functional, non-expiring Sybase IQ Developer's Edition for FREE.
Regards,
Bill
