To: investorgal who wrote (9764 ) 3/2/1998 10:03:00 AM From: Mark Finger Read Replies (1) | Respond to of 14631
>>If a customer is running the base ODS product and adds the extended >>parallel option for all their users then they are in for a heck of a >>surprise when none of their OLTP features work anymore. There are a few problems wrong with the above statement. 1. Virtually no OLTP systems run on MPP machines where XPS is commonly sold for data warehousing purposes. In general SMP machines have sufficient capacity for OLTP, but not for the bigger warehouse applications. 2. Clusters are more likely to be used for very large OLTP systems, not MPP. That is because clusters generally use a "shared-disk" architecture. Many MPP architectures are either "shared-nothing" or some hybrid. Because clusters have "shared disk", fail-over in a cluster is easier or simpler (than in most MPP cases). The basis for this is rather technical. Oracle has a "shared disk" for Oracle7, and is trying to move to better handle "shared nothing". IFMX XPS was developed for "shared nothing". Note that the recent record setting 1T TPC/D benchmark was run on a clustered system. 3. IFMX was supposed to add OLTP to XPS in the 8.2 version. Can anyone verify this? Something was improved, because to run the multi-user version of TPC/D, full ACID support (the technical name for proper transaction support) is required, and the "update" benchmarks must be run parallel to the general benchmark stream. 4. Generally if the XPO option is taken, only warehousing is done on that server. Usually the engine is tuned for either warehousing or OLTP operations anyway, and it is generally suggested that engines not run "mixed" loads (although this is supported by IFMX). One other thing to note is that this is not uncommon in the industry. For example, Sybase System 11, (IQ option) is even less capable of running OLTP than IFMX XPO. In other words, the previous posts comment are fundamentally a non-issue. Mark