I’ve been using the very useful scripts from FlashDBA to run SLOB2 on our new system, but unfortunately the analyze one is not RAC aware, so I’ve modified it, in very minor ways, such that it can use an AWR Global report (
awrgrpt.sql) as input and still extract the same values that the original does.
I call the script slob2-rac-analyze.sh
Here is an example run – ignore the numbers as they are not representative of anything in particular.
a555.net(jeff.a1):/app/support/SLOB: ./slob2-rac-analyze.sh rac_awr_12jul2013/awr.20.032/awr.20.032.txt > slob.csv
Info : Analyzing file rac_awr_12jul2013/awr.20.032/awr.20.032.txt
Info : Filename = awr.20.032.txt
Info : Update Pct = 20
Info : Workers = 032
Info : Read IOPS = 85.8
Info : Write IOPS = 33.0
Info : Redo IOPS = 15.6
Info : Total IOPS = 134.4
Info : Read Num Waits = 712
Info : Read Wait Time = 0.58
Info : Read Latency us = 814.606
Info : Write Num Waits = 926
Info : Write Wait Time = 0.28
Info : Write Latency us = 302.375
Info : Redo Num Waits = 2043
Info : Redo Wait Time = 0.37
Info : Redo Latency us = 181.106
Info : Num CPUs = 384
Info : Num CPU Cores = 192
Info : Num CPU Sockets = 24
Info : Linux Version = Red Hat Enterprise Linux Server release 6.3 (Santiago)
Info : Kernel Version = 2.6.32-279.2.1.el6.x86_64
Info : Processor Type = Intel(R) Xeon(R) CPU E7- 2830 @ 2.13GHz
Info : SLOB Run Time = 300
Info : SLOB Work Loop = 0
Info : SLOB Scale = 10000
Info : SLOB Work Unit = 256
Info : SLOB Redo Stress = LIGHT
Info : SLOB Shared Data Mod = 0
Info : No more files found
Info : =============================
Info : AWR Files Found = 1
Info : AWR Files Processed = 1
Info : Errors Experienced = 0
Info : =============================
Jonathan Lewis has a nice article covering the different AWR Reports.
I’ve only tested it on the system at work and it seems to work OK – your mileage may vary and I’d be happy to hear comments to the contrary, in relation to the changes I’ve made for use on RAC, but obviously the script is still 99% unchanged, so please contact FlashDBA if there are any generic issues you want to raise.
I’m not a unix shell script guy, but it seems to work…see what you think.
I enjoyed a trip out of the office today with my manager. We went down to the Oracle Enterprise Technology Centre, in Reading, to hear about the Oracle Optimized Warehouse Initiative. It was basically a half day pitch from Oracle and Sun today, although there are other dates with different hardware vendors (IBM, HP and Dell).
It was an interesting day, although I didn’t really hear anything new, per se. I think the main things I took away from the session were:
- The availability of a set of reference configurations providing matrices which cover various permutations of user count, storage cost focus, warehouse size and hardware vendor.
- The possibility of using the reference configurations either minimally, to simply cross check a proposed hardware specification for a given workload, to ensure it seems appropriate, or going the whole hog and using an “out of the box” reference configuration, delivered to your office, fully configured with all software installed, in a box, ready to run in your data.
- Oracle are pushing RAC and ASM heavily in the DW space – no surprise there.
- HammerOra and ORION are used by Oracle and the hardware vendors to assess the reference configurations…and there is nothing stopping you using them for your own benchmarking efforts
It was interesting to hear about the Proof Of Concept facility that Sun has, in Linlithgow, Scotland. The facility allows Sun (and an Oracle customer) to take a server off the production line and, working with the customer, test their workload on that machine to see if it’s going to work or not. Neat, and since we’re going to be using some Sun kit for a DW shortly, it sounds like an opportunity.
Funniest thing of the day for me, was the last slide in the pitch given by Mike Leigh of Sun which had the title “The Power of 2” and was illustrating the benefits to customers of Oracle and Sun as a united force. I didn’t really take much notice, as I was too busy smiling, as I looked at the title and it made me think of Doug and his Parallel Execution and the ‘Magic of 2’ paper (the Magic of 2 bit actually being from this paper by Cary).
If you’re building a warehouse, or just want to get an idea of whether your hardware is appropriate for the job, it’s probably worth reading up on the OWI.
I watched one of the presentations of the Audio CD from UKOUG Conference 2005 – Getting the most of of RAC on Linux by James Anthony and was interested to see his comments that reducing the block size / reducing the rows per block can help in tuning the interconnect between the instances of a RAC environment.
Basically, James was saying how reducing the number of rows per block, perhaps by reducing the block size or using a higher PCTFREE value, would help to reduce block contention at the cost of using more space. This would in turn make full scans slightly worse but he suggested that full table scans can be hard on a RAC environment anyway and are therefore best avoided….so I’m wondering how a RAC approach fits with a data warehouse, where full scans are reasonably common (albeit with some degree of partition pruning hopefully!) and also where we might use compression, PCTFREE 0 and large block sizes to maximise the rows per block.
It sounds like there might be elements of RAC that don’t fit well with warehousing – but I’m guessing it’s not necessarily black and white and that things can be designed/planned/managed appropriately.
We don’t use RAC on the system I’m working on currently and I’m not particularly experienced with RAC, so if anyone has any opinions I’d be interested to hear them.