Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: Joe NYC who wrote (76456)	4/4/2002 11:26:29 AM
From: pgerassi	Respond to of 275872

Dear Joe: Ever here of distributed database management systems? That is where the each piece of data is stored in more than one location. The typical setup involves storing a row in three separate nodes. That way if a node goes down and a node is in a scheduled outage for maintenance or upgrade, there still exists one copy so that operations may continue. Of course the easiest to program would be three identical systems each running the same RDBMS with all of the data with a small process making sure that all three are synchronized. The next stage is to have more than 3 servers and distributing the tables around so that no server contains more than 3/Nths of the data where N is the number of servers. Ideally, each server also handles the same number of operations. Realistically, this varies about 33% around the ideal case with smart planning can tighten. Given row level locking even tables can be distributed in pieces to the servers so those high transaction/size tables are evenly scattered among the nodes. This method scales with the number of nodes well (more than a super SMP box), is vastly more reliable and are used in the largest servers because they are not SMP boxes, but ccNUMA boxes. A ccNUMA DRDBMS (Distributed RDBMS) could have survived something like the WTC collapse, if enough nodes were remote to the WTC area (fiber ethernet can be done to a few km thus Beowulf clusters can be built from nodes in different buildings). A WTC collapse could take out 50% of the nodes, but the cluster and the DRDBMS keep going although somewhat slower making a $10 million an hour disaster to a $10 thousand dollar an hour inconvenience (if that). One such disaster could easily pay for many large DB servers. And that is far higher reliability than any SMP box could do. That is why most top end servers are ccNUMA type clusters rather than SMP boxes. And only the DRDBMS needs to know it is on a cluster as the API stays the same to the applications. This is what Oracle, Informix and IBM have in their flagship RDBMSes. Pete