The black box approach

From: Mladen Gogala <gogala.mladen_at_gmail.com>
Date: Sat, 17 Dec 2016 04:26:30 -0000 (UTC)
Message-ID: <o32epm$6q7$1_at_dont-email.me>



The time has come when Oracle application tuning is quickly becoming too complex to do efficiently. There is the basic optimizer, dynamic sampling, then +ACI-adaptive mayhem+ACI as J. Lewis has succintly put it, parallel execution and pifalls with direct read vs. buffered read. To add insult to injury, there is a whole lot issues with latches and mutexes, NUMA architecture and a whole set of new options like in-memory option and advanced compression options.
Of course, since all of that is not complex enough, the database can now run +ACI-in the cloud+ACI, so IO times and CPU times may not be what they seem. There is not much difference between random access IO and sequential IO in the cloud, because VMDK disks are essentially files, which may be fragmented. The result will be turning many of a sequential scans (scattered reads, as Oracle calls them) into random access IO, which is, of course, significantly slower than sequential access, when more than a single block is being read.
[Quoted] I am wondering whether it's time to take the +ACI-black box approach+ACI and use just the automatic tuning tools. The whole thing is overwhelming and to master it takes years. And when you do, it all goes invalid with the new, even more complex version.
[Quoted] At the same time, the relevant information is getting more and more scant and hard to find. Oracle is getting more and more closed every day.
-- 
Mladen Gogala
The Oracle Whisperer
http://mgogala.byethost5.com
Received on Sat Dec 17 2016 - 05:26:30 CET

Original text of this message