It was a freezing cold Birmingham that greeted UKOUG delegates on Monday to the start of the Annual UKOUG Conference. After hearing about lots of people having difficulty in getting to the conference, with some people even being snowed in making travel impossible, I felt a little smug that travelling from Oxford was such a breeze.
The first presentation I rolled up for was:
Sane SAN 2010 – James Morle
This was an outstanding presentation – very presentation zen like with illustrative pictures giving an idea of the concept of what james was saying, rather than screeds of text to read out.
This was basically a tour d’horizon of the state of storage
Obviously one of the big changes over the last few years has been SSDs, and James interesting view of the performance of SSD – he thinks it’s a game changer.
James thinks the days of Fibre Chanel over. In the past he advocated it over NAS and ethernet due to better performance (also stating the lack of good/reliable NFS clients), however for majority of use cases thinks that NAS and (10Gb) ethernet are more than adequate. On the high end James thinks Infiniband and SSD is the way to go. Interesting graph showing the latency of fc compared to infiniband – when using SSD the latency of the connection becomes a big factor, too big for Fibre Channel but acceptable for infiniband.
Oracle “Hidden” Features You Might Not Know About – Graham Wood
This was an interesting look at some new features that are “hidden” in the sense that there is not much razzmatazz around them and no Oracle marketing of them. some of them were incremental changes like:
FORMAT parameter can be ADVANCED – not documented
The talk was full of little snippets that were certainly new to me, I did not know you could set a parameter in a hint for a SQL statement using OPT_PARAM(…)
or that you can ignore embedded hints with parameter _optimizer_ignore_hints. In fact when upgrading Maria Colgan was advocating testing your applications with this parameter set as you may well have find that hints coded for several versions ago are no longer helping!
Pending statistics separates the collection of statistics from the publishing of statistics. This captures stats as pending allowing you to test that these statistics will be ok for your system. You can then publish them once verified.
You can load from an external table that is a gzipped file. You can pass a prepocessor argument in the create external table statement. You then use create table as… from the external table. Very easy to parallelize
It is possible to increase the number of top sql statements captured by AWR via dbms_workload_repository
AWR SQL report – reports history of execution of a SQL statement – useful if a statement is suddenly going slowly AWR RAC report shows load across different instances.
All in all this was a very useful talk where I picked up a lot of things that I was not aware off.
11gR2 Clusterware in depth – Frits Hoogland
This really was an in depth look at 11gR2 clusterware. I picked up a few things, and one of the best bits about this presentation were the dynamic slides showing the startup sequence of clusterware very effective way of explaining what in the Oracle manuals is a very complicated and unhelpful diagram.
In the hc_sid.dat tells clusterware state of resources – clusterware will know if the db has been shutdown manually. I just checked an hc_sid.dat file and the strings output is quite amusing:
oracle@linuxrac1:DBA1 /var/opt/oracle/product/11.2.0/dbs> strings hc_DBA1.dat DO NOT DELETE OR OVERWRITE THIS FILE!!! DBA1
Really glad I went to Frit’s presentation.
Performance Stories From Exadata Migrations – Tanel Poder
I don’t see myself being involved with Exadata in any sort of way but this was Tanel Poder. It seemed a little of an unusual presentation from him, much more like a “War Story” of his involvement in some projects with Exadata. Was of course instructive and in particular the migration part whereby he went through how he investigated the issues with moving data onto the Exadata platform.
Optimizer issues – Maria Colgan
This was an excellent presentation by a fantastic speaker. Maria presented lots of examples and for an Oracle speaker was quite open about the product. I particularly liked here point about when you see factors of 1% or 5% you know the optimizer is just guessing.
dynamic sampling can kick in automatically in 188.8.131.52 when parallel query in use.
All in all a truly excellent day, where I picked up a lot of tips.