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 : How high will Microsoft fly? -- Ignore unavailable to you. Want to Upgrade?

To: alydar who wrote (63110)	11/16/2001 7:30:44 PM
From: rudedog	Respond to of 74651

Rocky - the transport I worked on was the VI interface - I got involved because I was one of the original VI design team in 1996-97. The ORCL work was quite recent and is part of the current product. I wrote some of the material in the Oracle white papers which are on their website. VI cut the overhead to move a lock from one global cache instance to another from 40,000 instructions to about 5,000 - a big enough difference so that the older DLM based OPS systems, which didn't scale at all, became scalable once the overhead to establish and maintain the "cache fusion" links was overcome. Your claim of 90% scalability is correct between 2 and 6 nodes. It takes some very special tweaking to have the relationship between 1 node and 2 be anything but mediocre - 2 node systems are really best suited for availability (which they do pretty well). And it rolls off pretty badly for larger numbers of nodes. The latest Oracle clusters are a generation ahead of OPS but the claim of unlimited scalability with today's technology is not realistic, and scaling beyond a handful of nodes is unlikely until infiniband is widely deployed. I wish you the best of luck with your customers.