Storage Replication is For Losers – James Morle
I was a bit late out of bed, so I had to hurry to jump just in time in the right room. I had decided to follow James Morle’s session about data replication. James had just become father again, so we were treated with some pictures here and there in the presentation.
But what it really was about, was that the usual storage replication does things in the wrong order and can make the wrong things wait. James warned us of consistency groups: storage grouping of database files that force data to replicate in order, even though you want the redo to go first. What you get, is longer log file parallel write times, sometimes in seconds. (Mental note: check if we have these problems. Then visit storage team)
James’ argument is that Data Guard does know what needs to go first, so it’s a better way of replication. He also mentioned a new feature in 12c: far sync, where you can send arch and redo data to a far sync point, from where you can sync to several databases. There was a session later in the day that would tell more about this, but ..you know.. choices.
The Least an Oracle DBA Needs To Know About Linux Administration – Martin Nash
Martin Nash discussed a lot of commands that the DBA needs to know about. His sheets mainly consisted of lists of useful Linux commands, and he told why we needed to know about them. I learned that I should use ip instead of ifconfig, that dig is better than nslookup, that in Oracle Enterprise Linux there is a WHENEVER SQLERROR command (but don’t remember where exactly).
I already learned to like the screen command. But apparently you can give access to your screen commands to a collegue and I just catched the phrase “security risk” concerning that. On RAC the watch command is useful to see what’s going on on a cluster.
I already knew interesting features of the top command. I didn’t know you can filter on a username with the u key.
All About Joins – Tony Hasler
This session was in the illustrious room Ex4 (and 3). Here you get two presenters for the price of one, because the sound from the other room isn’t properly isolated. There is only a curtain between the rooms and that is not nearly enough.
So I had a hard time following Tony’s session and it wasn’t really his fault. At least I could make out that Tony advices us to learn ANSI joins, because they are much clearer formatted and you can tell more clearly what you actually want to join on what.
12c sees the introduction of the lateral join (select * from t1, lateral (..)), but actually I’ll have to read the presentation back to see what that was all about.
There was much more, but I’ll have to read more into this to get the meat of the session. Or read his book. You can one at this conference BTW.
Oracle 12c Database in the Enterprise Cloud – David Hickson
I went to this session, because one of our infrastructure architect asked me to get more information about databases in the cloud. And that was actually a good idea, because David Hickson had lots of information to share, based on his experiences at British Telecom.
He told about how at BT already 674 databases were deployed in just one year: XML driven and with Python (apparently Julian Dyke told about a similar methodology in the session before this one).
David told about how Cluster Health Monitor wants to use the “first disk group” to store it’s data and that this is usually the vote disk group. There is not only not desirable, but also there is no way around it, except a workaround lateron.
He told about Flex ASM (in where you don’t need an ASM cluster on every node) and that they are talking with Oracle about whether this is the “future” or if the old method is. And he told about how you can maximize the maxsize of tablespaces in pluggable tablespaces, but the way around that (still available) is just to add more datafiles. v$pdbs needs a couple of seconds to realize what has happened.
He talked about disabling features, like Active Data Guard (_query_on_physical=false) and I just catched that this has consequences for block change tracking.
ASM Metrics Are a Goldmine – Bertrand Drouvot
Bertrand Drouvot started with what asmcmd iostat can do for you. It’s actually not that RAC aware. There is also gv$asm_disk_stat that gives I/O stats, but not per disk. And in 12c there is now gv$asm_disk_iostat that does have the I/O stats per disk. Bertrand told us about asm_metrics.pl, a script that (I think I understood) is in the Oracle software tree.
Then he did a daring thing: he did a live demo and not just a script that ran everything from start to finish. He also showed how to get the data out as .csv and use it in Tableau, a graphing tool that’s rather popular nowadays. (Sometimes it’s not just Oracle products that you learn about on conferences like these.)
In the evening I had diner with collegues of my old employer, Transfer Solutions. After our diner talk, I concluded we must write a book of strange tales from IT, because we just have too many of them.