Re: [Iometer-user] Strange (Impossible) Results, and PerformanceCounters (???)

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Thanks for you response. Can you tell me what type of cpu they uses? and
the ghz? I have a centrino one and a pentium 4 Mobile here. maybe i can
reproduce the error.

i will try to give u a debug version as soon as i find a visual c++. :P.

ming

On Mon, 2004-03-15 at 14:19, Kenneth Morse wrote:
> To clarify your clarifications:
> 
> > 1) this is related with some computer hardware and happen under both
> > windows 2k and xp. and have nothing to do with drives.
> 
> Correct.
> 
> > 2) and when this improper results shows, there are some print out like
> > '"Performance Counters greater than 1" followed by increasing number'.
> > or
> 
> More often it's negative numbers, but positive large numbers happen as well.
> Note that I can not 100% conclusively tie this error as emitted by IOMeter
> to the strange results.  They *seem* to go together, but that does not
> necessarily prove causality.
> 
> > 3) all these problem system are ibm t23 and r31.
> 
> All T23s and R21s that I have tried exhibit the problem.  By no means does
> that mean all T23s or R31s are bad - or that the problem is confined to just
> these two models.
> 
> > 4) with pm enable, the system perform well while with pm disable, the
> > systems go wrong.
> 
> Nope, let me rephrase:
> 
> On all systems that do not give me problems (aka Good Systems), Iometer
> works correctly regardless of the PM setting.  Obviously with PM enabled,
> when the machine throttles back the CPU, it does have an effect on both CPU
> utilization and performance, but IOMeter reflects these correctly.
> 
> On the systems that DO give me problems (aka Bad Systems), Iometer does not
> appear to report numbers correctly regardless of whether or not PM is
> enabled.  I have PM disabled just to rule it out as a source of the
> problem - but there is no setting that I can find yet on the Bad systems
> that will make IOMeter behave properly on them.
> 
> I will gladly try out a Debug build.
> 
> Thanks,
> 
> -Ken
> 
> 
> ----- Original Message ----- 
> From: "Ming Zhang" <mi...@el...>
> To: "Kenneth Morse" <kd...@be...>
> Cc: "iometer-user-list" <iom...@li...>
> Sent: Monday, March 15, 2004 9:32 AM
> Subject: Re: [Iometer-user] Strange (Impossible) Results, and
> PerformanceCounters (???)
> 
> 
> > i try to summarize u description in order to better understand your
> > question.
> > 1) this is related with some computer hardware and happen under both
> > windows 2k and xp. and have nothing to do with drives.
> > 2) and when this improper results shows, there are some print out like
> > '"Performance Counters greater than 1" followed by increasing number'.
> > or
> > 3) all these problemful system are ibm t23 and r31.
> > 4) with pm enable, the system perform well while with pm disable, the
> > systems go wrong.
> >
> > i think it is hard for us to debug this since we do not have such
> > laptop. my intention is to give a iometer version that has some debug
> > information and then please run it on u system and give us some
> > feedback. do u think so? thanks.
> >
> > ming
> >
> > On Mon, 2004-03-15 at 01:38, Kenneth Morse wrote:
> > > I have been running a large number of iometer tests across a large
> number of
> > > drives over the past two weeks.  Tests have included block sizes ranging
> > > from 512B to 1MB (Exponentially), IO Depths ranging from 1 to 4096, and
> > > several access patterns including full sequential read, sequential
> write,
> > > random read, random write, and many combinations.  All in all, over 1000
> > > tests performed on each drive.  The results have been fascinating, and
> about
> > > what would be expected.
> > >
> > > For example, the transfer rates on Sequential Reads grew significantly
> from
> > > poor performance using 512B transfers to peak performance around 16KB
> > > transfers.  The performance from 16KB up to 1MB and beyond was fairly
> > > constant, and was very close to the rated sustained transfer rate of the
> > > drive in question.  Performance of course varied based on what part of
> the
> > > drive was being accessed, etc....
> > >
> > > Small block random performance varied based on drive speed, seek times,
> > > drive size, etc... again, all exactly as expected.
> > >
> > > However, on a handful of systems (and this is system related,
> independent of
> > > the Drive installed), I am receiving impossible and inaccurate results.
> I
> > > have reinstalled the systems, tried both W2K and XP, along with
> everything
> > > else I could think of.  On these systems, ALL results from ALL tests
> seem
> > > inflated by a factor of 6-10x.  For example, a drive that has a maximum
> > > sustained transfer rate of around 34MB/s, is clocking in at 300MB/s
> under
> > > IOMeter.  (Sequential read, 16K to 1M blocks, 1 I/O, testing entire
> drive).
> > > At the same time, if I bring up windows performance monitor, it shows a
> > > constant data rate of 33MB/s to the drive.  On the systems that are
> > > behaving, windows performance monitor and IOMeter agree nearly
> perfectly.
> > > On these flaky systems, windows shows exactly what I expect during an
> > > IOMeter run, and IOMeter shows and records impossible values.
> > >
> > > After a week of tinkering, I am at a complete loss.  I have tried every
> > > (stable) version of IOMeter here on sourceforge with the same results -
> > > going all the way back to Intel's original release.
> > >
> > > If I look at dynamo.exe while this is happening, I sometimes (but not
> > > always) see information on the order of "Performance Counters greater
> than
> > > 1" followed by increasing numbers, 1.1, 1.2, 3, 5, 9.  Sometimes I see
> > > "Performance Counters below zero", followed by a list of negative
> numbers.
> > > I suspect this may have something to do (be the cause of) with an
> improper
> > > calculation of MB/s (and all other * per second numbers), but I don't
> know.
> > >
> > > Many of the machines involved are IBM Thinkpads.  For example, I have
> here
> > > well behaved T20s, T21's, T40's, and T41's.  And I have a T23 and a R31
> that
> > > are constantly out of step with reality by a factor of 6x-10x.  All
> power
> > > management features have been disabled.  (And as a point of interest,
> when
> > > PM does kick in and throttle the processor or the buss, performance gets
> > > worse, and IOMeter reports is correctly.. it even reports a throttled
> back
> > > Mhz rating - slick!).
> > >
> > > I am baffled, and completely at a loss here.
> > >
> > > -Ken
> > >
> > >
> > >
> > >
> > >
> > >
> > > -------------------------------------------------------
> > > This SF.Net email is sponsored by: IBM Linux Tutorials
> > > Free Linux tutorial presented by Daniel Robbins, President and CEO of
> > > GenToo technologies. Learn everything from fundamentals to system
> > > administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
> > > _______________________________________________
> > > Iometer-user mailing list
> > > Iom...@li...
> > > https://lists.sourceforge.net/lists/listinfo/iometer-user
> >
> >
> >
> > -------------------------------------------------------
> > This SF.Net email is sponsored by: IBM Linux Tutorials
> > Free Linux tutorial presented by Daniel Robbins, President and CEO of
> > GenToo technologies. Learn everything from fundamentals to system
> > administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
> > _______________________________________________
> > Iometer-user mailing list
> > Iom...@li...
> > https://lists.sourceforge.net/lists/listinfo/iometer-user