Comments on: Oracle “Physical I/O” ? not always physical

By: Jagjeet Singhh

Jagjeet Singhh — Thu, 02 Aug 2012 21:20:05 +0000

Hi Kyle- Thanks for sharing, you have mentioned “How can we tell the difference between a read from the underlying storage or a read from the O/S file system cache? ”
Does that (file system cache ) apply even when we use ASM ? Please clarify

By: Kyle Hailey

Kyle Hailey — Wed, 25 Jul 2012 18:12:04 +0000

Hi David,

Thanks for adding parsetrc.pl to your page on trace parsers and thanks for the page in general. I’ve found it helpful.

RE: github – yeah, github looks awesome. Actually it looks so nice it’s hard to believe. Looking forward to using it more. Also https://gist.github.com looks nice for sharing code in URLs and blogs

- Kyle

By: David Mann

David Mann — Tue, 24 Jul 2012 17:18:26 +0000

Looks like I have another entry for my page :) I seem to get a lot of hits on my 10046 tool page so I gave it a pretty alias and will try to keep it up to date. http://ba6.us/oracle_sql_trace_analysis_tool_list

I am in the middle of the same sort of investigation but unfortunately my Prod is much slower than Dev… even with the same cloned data behind them. I have traces collected and was going to make my way through the list… Thinking I will try your script and OraSRP first.

And thanks for putting your stuff out there on GitHub. I just started shovel… errr moving some of my old code out there as well. I have some Java 10046 parsing code that I was working on that I need to tidy up and make available. Until I can get my company to spring for MethodR tools I will have to rely on my own code and the kindness of folks that release their code into the wild.

By: Kyle Hailey

Kyle Hailey — Thu, 19 Jul 2012 00:12:57 +0000

Looking at OraSRP from
http://www.oracledba.ru/orasrp/sample3.html
it looks like the current sections are:

Summary (dbid,node,sid,vers,#SQL[#user,#recur],#trx[#rollback,#readonly]
Session Flat Profile (Top Event breakdown)
Top 5 Statements per Event ( for each event list the top sql for that event)
Session Call Graph ( graph of recursive SQL)
Events Histograms
SQL Statements
Statement Cumulative Statistics
Statement Self Statistics
Statement Flat Profile (top events)
Statement Plan
Statement Binds Statistics

Good stuff.
Session call graph is nice as is organizing Top SQL for each top EVENT.

The example http://www.oracledba.ru/orasrp/sample2.html
shows I/O latency by I/O size which is nice.

The event histograms and the latency by I/O size are the two main things I was addressing parsetrc.pl. Some other things that are in parsetrc.pl that are still being polished are MB per second, max and min I/O throughput of any single I/O grouped by second broken out by I/O type – sequential, scattered, direct path, direct path temp.
Also I/O rates per minute.
Also putting together are compact table of SQL with just the SQL_ID the PLAN HASH and elapsed, cpu, wait, unaccounted for, ordered by elapsed as well as CURSOR_ID (the ID used in the trace file)

By: Kyle Hailey

Kyle Hailey — Wed, 18 Jul 2012 14:26:57 +0000

@Timor,
Thanks for the note on OraSRP. I hadn’t realized that gave histograms now. Here is a list of 10046 parsing tools out there: http://ba6.us/node/177
Cary’s Method-R tools are the best by far but of course they are pay for tools.
OraSRP looks fantastic. I haven’t used it in ages. One thing I liked about OraSRP was trying to distinguish idle SQL*Net message from client waits from SQL*Net message from client waits that occur between fetches and are a more clean indicator of some application issue. Unfortunately OraSRP is only available on some platforms and the source isn’t available.
I like having scripts because it’s nice to be able to modify and with scripts the should/could run anywhere.
- Kyle

By: Timur Akhmadeev

Timur Akhmadeev — Wed, 18 Jul 2012 10:01:50 +0000

Hi Kyle

nice article, thanks. I want to add a link to OraSRP which can produce similar histogram output. Here’s a text example (html is the default with a graph):
[sourcecode]
db file sequential read
———————–
value ————- Distribution ————- count
0.000004s | 0
0.000008s | 15
0.000016s |@@@ 113
0.000032s |@@@@@@@ 204
0.000064s | 6
0.000128s | 26
0.000256s |@@@@@@@@@@@@@ 387
0.000512s |@@@ 96
0.001024s |@ 45
0.002048s | 10
0.004096s | 27
0.008192s |@@@@ 132
0.016384s |@ 54
0.032768s | 10
0.065536s | 5
0.131072s | 1
0.262144s | 0
[/sourcecode]

By: François GREGOIRE

François GREGOIRE — Wed, 18 Jul 2012 07:51:26 +0000

Nice stuff, it makes me think that I have to look into this R thing, it might save me hours of painfull discussions with my system admin…gosh I was one of them before, now I can’t suffer them any more…:)