Oracle Parallel Execution: Interconnect Myths And Misunderstandings

Oracle Parallel Execution: Interconnect Myths And Misunderstandings

Benchmark IO

Windows ORION Numbers

Oracle Orion

Prompted by Jeff Moss’ blog, I decided to use ORION to measure the performance of various I/O configurations I’m testing for the upcoming Hotsos presentation. I’ve already blogged about the Linux results on a couple of configurations but I’m still tempted to include a Windows Laptop test (although I haven’t decided yet). To that end, I’ve been playing around with the Windows version of ORION today, using XP Professional, SP2.

I ran into a few problems. The first was that, like Jeff, I didn’t have any raw partitions or volumes for the test and didn’t fancy creating any on my main laptop. So I was restricted to testing on files. I initially tried to test a collection of eight Powerpoint files, so my simple.lun file looked like this. (The files themselves are just a variety of presentations)

C:\TEMP\1.PPT
C:\TEMP\2.PPT
C:\TEMP\3.PPT
C:\TEMP\4.PPT
C:\TEMP\5.PPT
C:\TEMP\6.PPT
C:\TEMP\7.PPT
C:\TEMP\8.PPT

I picked Powerpoint files because Jeff had found that he could only use MS Office files and these were some of the biggest. I ran the test using the following command. (In fact, this is the command I used for all of the following tests – I just edited the contents of simple.lun each time)

orion -run simple -testname simple -num_disks 1

and it seemed to work okay for a few minutes before failing with the following error.

ORION: ORacle IO Numbers — Version 10.2.0.1.0
Test will take approximately 9 minutes
Larger caches may take longer

storax_aiowait: IO returned an error 27070
lun_aiowait: storax_aiowait failed.
rwbase_run_test: rwbase_reap_req failed
rwbase_run_process: rwbase_run_test failed
rwbase_rwluns: rwbase_run_process failed
orion_thread_main: rw_luns failed
Test error occurred
Orion exiting

I think the 27070 is a generic Windows i/o error. By a process of elimination (otherwise known as faffing around) I managed to get it to work with only one file, with the following results.

Maximum Large MBPS=18.27 @ Small=0 and Large=1
Maximum Small IOPS=231 @ Small=2 and Large=0
Minimum Small Latency=4.43 @ Small=1 and Large=0

Which are stunning. So good, in fact, that I didn’t believe them for more than a few seconds. The hard disk in a cheap laptop is unlikely to be producing nearly double the number of IOs per second of even a previous generation U160 SCSI drive :-

Maximum Large MBPS=25.65 @ Small=0 and Large=2
Maximum Small IOPS=138 @ Small=4 and Large=0
Maximum Small Latency=8.92 @ Small=1 and Large=0

Then it occurred to me that testing a tiny 2Mb file is hardly a useful test of drive performance. Looking around for a bigger file, I figured I should try one of the database files!

C:\oracle\product\10.1.0\oradata\orcl\TEST_DATA01.DBF

Which worked fine and seems like the best type of file to use, particularly if I had something bigger than this 3Gb example. Here are the final results.

Maximum Large MBPS=17.52 @ Small=0 and Large=1
Maximum Small IOPS=79 @ Small=5 and Large=0
Minimum Small Latency=14.21 @ Small=1 and Large=0

They look more sensible to me when compared to the (old PC) single IDE Linux example

Maximum Large MBPS=14.46 @ Small=0 and Large=1
Maximum Small IOPS=46 @ Small=5 and Large=0
Maximum Small Latency=22.96 @ Small=1 and Large=0

I still think the utility is likely to work more effectively on unused drives rather than a few files.

http://download.oracle.com/otn/utilities_drivers/orion/Orion_Users_Guide.pdf

Tuning Oracle RDB for DWH purposes

Для меня эта тема пока является открытым вопросом – поэтому я завел тред’s на sql.ru – http://sql.ru/forum/actualthread.aspx?tid=657735 и на hyperion.su - надесь что с помощью коллективного разума сумею добраться до истины.