Dude, you rock.

Happy 2008.  I need to resolve better.

:-)

Chuck

That's my home town!

That's a decent read, but being from 1999, how much of the methodology and conclusions do you think are still valid with today's processors?

Yeah, "Pentium II Xeon processor running at 400 MHz, with 512 MB of main memory

connected to the processor chip through a 100 MHz system bus" is hardly a modern system these days. But I find the discussions overall interesting. More specifically, its choice of the measurement tools, its discussions of the query execution model, and its discussions of the query execution time breakdown are quite relevant (to my job at least), though the specific numbers have changed or may not apply to the new processors. Speaking of the newer processors, I hope the author would repeat the experiment on a modern multi-core processor.

Also, the paper is still widely quoted in the recent literature.

Linchi,

Thanks for sharing the link to this paper.  The findings are interesting and relevant despite the time that has passed since its publication.

My gut feel says that while current systems are technically "faster" (higher clock speed, faster cache, faster bus, faster memory, multi-core, etc.) and theoretically “more powerful” than the systems tested in the paper, the problems highlighted by the paper may actually be more acute now than they were then, since the blocking / stalling issues are brought about by differences in the fastest and slowest paths for a given operation and these differences may be proportionally greater now than they were then.  As a result, the percentage of system resource available to do useful work may be less now than in the earlier test results.  The large overall amount of capacity and performance growth in current systems versus the tested systems still yield greater performance, but the amount of gain may be diminishing due to this effect.

Again, just a gut feel on my part – no specific evidence to back it up.  Like you stated, I think it would be great if the tests could be repeated on current systems, but I know the effort involved in such a test would be pretty large.

I would be interested in your opinion of how current systems might compare to the test results and findings in the paper.

The paper refers to “database developers”, which I take to mean product development staff for RDBMS vendors, and not IT application developers or DBAs that use an RDBMS product.  In my opinion, it is only the vendor product development staff that will be able to change these behaviors (stalls, blocking, etc.) due to code and data caching issues, etc.  I would be curious to know how much of this type of work (make the current functions perform much better through carefully analysis and testing) is actively done by RDBMS vendors to improve their respective products, versus just expanding the list of available features and capabilities to sell more product.  I suspect some is done, but how much and how deep (to the level of analysis done by the paper) I don’t know.

I am also curious as to which lettered system (A, B, C, D) corresponded to which vendor / RDBMS product / version.  I know that legal restrictions prevented them from revealing this information, but it would be interesting to know.  I’m guessing that the mix was the RDBMS market leaders at that time, and may have included Microsoft SQL Server (because Microsoft provided the OS for the test system(s) – probably 7.0 because 2000 had not yet been released in 1999), Oracle (probably 7 or 8), IBM DB2, and maybe Sybase.  I would have no clue as to which was which in the test results.  Any guesses on your part?

Thanks,

Scott R.

In my previous post on this subject , I mentioned that I'd very much like to see the authors repeat the