|
Archives of the TeradataForumMessage Posted: Thu, 07 Jul 2005 @ 10:16:53 GMT
Hi Charles, Teradata is an MPP share nothing architecture. It scales by adding extras nodes with all data being distributed across all nodes. The bynet is used to distribute result sets rather than the raw data. This architecture means that a Teradata system does not need a huge memory space to make data available across nodes. The NUMA technologies do this and suffer from the fact that non local memory access is slower than local so as a system scales and more data is not local performance falls off. There is less benefit and for Teradata in moving to larger nodes. The strategy is to move to 64 bit when it is cost effective to do so. Teradata is running already on 64 bit on non NCR hardware. This makes sense as without the bynet these systems can not achieve MPP scalability. The latest generation 5400 uses processors with Extended Memory 64 Technology (EM64T). EM64T provides the capability to run on current 32-bit architectures while providing future 64-bit capabilities. The idea is that these nodes can be upgraded to 64 bit when it is advisable to do so. Sorry if this has turned into a sales pitch I just like the technology. Daniel O'Hara
| ||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||
Copyright 2016 - All Rights Reserved | ||||||||||||||||||||||||||||||||||||||||||||||||
Last Modified: 15 Jun 2023 | ||||||||||||||||||||||||||||||||||||||||||||||||