Re: SQL stuff...
Originally posted by MacCoaster
This [unisys.com] isn't scalable enough for you? 25k concurrent users, that's one of the records. 57,000 [microsoft.com] concurrent users!!! What about this [tpc.org]? You'll see Microsoft SQL Server in all top tens and some even blow away others.
Please back up your claims of Microsoft SQL Server having a poor reputation for scalability.
Hmmm, can't give any hard facts any more (at least, nothing you wouldn't be able to easily find yourself by going to Google and typing in "Oracle scalability"), as I'm not dealing with that these days. I can tell you that I worked on two major projects in the past five years that performed exhaustive scalability testing of SQL Server (2000 and the version before that) with a scalable clustered (not single-machine as your two links point to, which is largely unheard of in major applications) hardware architecture. We found that as demand on SQL Server grew the hardware requirements grew disproportionately, and the "dead top" of scalability was fixed based on the power of the individual clustered machines. This was in striking contrast to Oracle, which scaled up to terabytes of database and hundreds of individual servers without the size/hardware and transactions/hardware graphs spiking up.
The number of concurrent users on a database is relatively meaningless. What is meaningful is the number of users that can be doing actual work on the database at one time. This implies a load analysis, not a "how many users can we log in without bringing down the system" mountain climbing exercise. This also implies you are working with a large amount of test data underneath, the size that would be generated by your thousands of concurrent users. In real world tests on two different real world systems, my teams found SQL Server to be sorely lacking in terms of high-end capacity.
The second important factor in scalability is flexibility: if your demand grows, can you handle it by just adding more hardware, or do you have to rearchitect software? Adding hardware, even at $100k/box, is much cheaper than the manpower required to rearchitect and tweak software. Again, this is a situation where SQL Server was found to be lacking.
This was not a foregone conclusion, either. Although both teams had fervent Oracle supporters (the ones who had been doing this long enough to notice that Oracle rarely loses such analysis), both teams also had people going in thinking SQL Server would come out on top, and financial pressures to use SQL Server (not the least of which were MS promised grants).
Competition is good for Oracle, and certainly good for its customers, but SQL Server still does not yet have a reputation for scaling well in the industry. It will need more than MS-sponsored "studies" to change this opinion.
As I said, however, these factors don't mean much considering that a DBFS is an entirely different beast than a production database, with different design constraints and different compromises to be made. The best server DB will not necessarily be the best DBFS. And, personally, I don't think anyone will be willing to pay millions of dollars for a ten-user license of Oracle on their desktop computer to do nothing more than serve them their files.
No matter what its successes and failures in the production arena, SQL Server has a completely different ballgame in backing up this "Windows Future Store" (ick) FS.
