Re: DBMS Selection for 500+GB data and 500+ Users

msigler@informix.com (Mark Sigler) writes: > >Don't be sucked in to oversizing the hardware. Oracle will tell you that >the only way to scale that large is with an MPP system. Funny that I've >done it with Oracle on large SMP systems from Sun and outperformed MPP. >which causes excessive disk i/o while trying to maintain cache consistency. So how can a Data Warehouse cause the caches of nodes in a cluster to need distributed locks? Are the same pages in different nodes being updated at the same time? no Is data updated in a data warehouse? no If your data warehouse is for DSS, then the number of distributed lock requests and pages passed between nodes is ZERO! There is no cache consistency issues when multiple nodes perform reads of the same data. Oracle does not have read locks and nodes only use the distributed lock manager (DLM) when trying to perform an update to a page (to make sure that page does not exist in another node's cache). Since pages are read and written to disk (2K, 4K, 8K, and 32K on OSF/1), it would be a bad thing if a user on one node updated row 10 on page 10 and another user updated row11 on page 10 (from different nodes). To avoid lost updates, a clustered system (Informix has none), uses the DLM to maintain cache coherency. Clusters are a "future" announced from Informix. When they get there, and if they support updates, they will have to use a DLM to manage cache coherency. You point is not correct. Regards, -=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()} David Natelson Oracle Corporation dnatelso@us.oracle.com -=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()}>]:=--=:[<{()}