Re: access IO by M/M/m Module

From: Cary Millsap <cary.millsap_at_method-r.com>
Date: Tue, 27 Sep 2011 08:52:14 -0500
Message-ID: <CAJOkrQYmF50HE+wq0p1qNfye4kbZMd686JQUeTjCCH8tK9kavQ_at_mail.gmail.com>

"db file sequential read" call durations reported by Oracle are not service times (S); they are response times (R), which include both service time and queueing delay (S + W). They are not suitable inputs into the S slot in an M/M/m queueing model.
Cary Millsap
Method R Corporation
http://method-r.com

On Mon, Sep 26, 2011 at 10:17 PM, Sidney Chen <huanshengchen_at_gmail.com>wrote:

> Hi,
> When reading the Topic of M/M/m Module in the book <<Optimizing Oracle
> Performance>> by Gary Millsap, I have the idea to verify if our I/O
> subsystem can fit the M/M/m module.
>
> The db is IO intense, 80% db time is spent on "db file sequential read",
> and the avg wait time is 15ms.
> 1. to do the assessment, I extract 3357 lines in one 10046 trace file
> 2. compute the ela time as second(divide by 1000000, keep the distribution
> compact)
> 3. using the mdist.pl to tell if the elapse time for "db file sequential
> read" fit the "exponential distribution".
>
> the result is reject, so i wonder if there are accurate methods to assess.
> Any advise is appreciated.
>
> oracle: ~/scripts$ perl mdist.pl 2.lst
> Hypothesis: Data are exponentially distributed with mean
> 0.0155063946976467
> Test result: n=2 V=1143.25 p=1.000
> Verdict: *Reject*
>
>
>
>
> --original record in 10046 trace file
> WAIT #19: nam='db file sequential read' ela= 26431 file#=207 block#=258517
> blocks=1 obj#=206447 tim=1284706983664908
> WAIT #19: nam='db file sequential read' ela= 11191 file#=108 block#=1351954
> blocks=1 obj#=206444 tim=1284706983677002
> WAIT #19: nam='db file sequential read' ela= 11836 file#=25 block#=294842
> blocks=1 obj#=206444 tim=1284706983689523
> WAIT #19: nam='db file sequential read' ela= 6268 file#=25 block#=472857
> blocks=1 obj#=206444 tim=1284706983696181
> WAIT #19: nam='db file sequential read' ela= 6339 file#=108 block#=1348802
> blocks=1 obj#=206444 tim=1284706983702664
> WAIT #19: nam='db file sequential read' ela= 7363 file#=25 block#=434781
> blocks=1 obj#=206444 tim=1284706983710368
> WAIT #19: nam='db file sequential read' ela= 12006 file#=176 block#=489651
> blocks=1 obj#=206444 tim=1284706983722477
> WAIT #19: nam='db file sequential read' ela= 18163 file#=107 block#=1728592
> blocks=1 obj#=206444 tim=1284706983740767
> WAIT #19: nam='db file sequential read' ela= 24003 file#=107 block#=1740869
> blocks=1 obj#=206444 tim=1284706983765194
> WAIT #19: nam='db file sequential read' ela= 4780 file#=25 block#=473096
> blocks=1 obj#=206444 tim=1284706983770782
> WAIT #19: nam='db file sequential read' ela= 6644 file#=25 block#=431224
> blocks=1 obj#=206444 tim=1284706983777508
> WAIT #19: nam='db file sequential read' ela= 20201 file#=107 block#=1747395
> blocks=1 obj#=206444 tim=1284706983797791
> WAIT #5: nam='db file sequential read' ela= 89939 file#=207 block#=359305
> blocks=1 obj#=206447 tim=1284706984136333
> WAIT #5: nam='db file sequential read' ela= 21708 file#=25 block#=292852
> blocks=1 obj#=206444 tim=1284706984158777
>
> --record in 2.lst, ela divided by 1000000
> .026431
> .011191
> .011836
> .006268
> .006339
> .007363
> .012006
> .018163
> .024003
> .00478
> .006644
> .020201
> .089939
> .021708
>
> --
> Sidney Chen
> Blog: *http://sid.gd* <http://sid.gd>
>
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>

--
http://www.freelists.org/webpage/oracle-l

Received on Tue Sep 27 2011 - 08:52:14 CDT