I have time to time brought in selective performance-enhancements into client situations. I even wrote a white paper about one of them, RightOrder. SAP NetWeaver BI Accelerator has come in and out of my life several times. Itâ€™s an in-memory analytic engine that performs its queries in-memory. It actually has quite a broad base of deployment and is shipped as an appliance (HP or IBM with Linux) and I think itâ€™s worthwhile to think about in selective situations. Itâ€™s a BI product for SAP NetWeaver. There are features of the product that add a lot of value, like the high compression, which is enabled by the column-wise ordering of the data, which Iâ€™ll discuss below.
There is also the horizontal partitioning across multiple machines which enables the solution to get beyond previous in-memory limitations. Itâ€™s optimized for various common data types and has different algorithms for different data types. I donâ€™t know exactly how much of the query performance improvement comes from this, but I suspect itâ€™s quite a bit. Architecturally, they put more memory in the cache â€“ another performance improvement technique.
The column-wise ordering of data allows it to perform very high selective compression because all of a columnâ€™s values are physically together. It also provides for excellent performance when you select a small subset of the columns in a table since you do not perform I/O for data that is not needed. Column-orientation greatly assists a compression strategy due to the high potential for the existence of similar values in columns of adjacent rows in the table. Itâ€™s ideal for columnar functions like SUM, COUNT, MIN, MAX, and AVG.
So, the shoe fits here when you need a specialist data mart useful when the performance of limited-column queries, especially columnar functions, is the overriding selection factor.
Posted May 9, 2007 5:06 PM
Permalink | No Comments |