Re: [dtrace-discuss] Is there a way to clear Dtrace arrays?
> Ignore my request. Even if there was a way to clear an array, there > is no function that allows me to print out that array. Only an > aggregation does that. True, and I'd argue that an aggregation is probably the right data structure for what you're doing. Adam -- Adam Leventhal, Fishworkshttp://blogs.sun.com/ahl ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Hello all... Thanks a lot for the answers! I think the problem is "almost" fixed. Every dtrace documentation says to use predicates to guarantee the relation between the start/done probes... Max was the only one paying attention reading the docs. ;-) But i'm still getting weird numbers: Wed Dec 10 08:36:33 BRST 2008 Wed Dec 10 08:46:55 BRST 2008 cut here - Tracing... Hit Ctrl-C to end. ^C NFSv3 read/write distributions (us): read value - Distribution - count 2 | 0 4 | 631 8 | 145603 16 |@155926 32 |@@ 15970 64 |@6111 128 | 942 256 | 372 512 | 883 1024 | 1649 2048 | 1090 4096 |@8278 8192 |@@@ 24605 16384 |@8868 32768 | 1694 65536 | 304 131072 | 63 262144 | 27 524288 | 31 1048576 | 43 2097152 | 3 4194304 | 0 write value - Distribution - count 128 | 0 256 | 1083 512 | 32622 1024 |@70353 2048 |@@ 70851 4096 |@@ 47906 8192 |@@ 44898 16384 |@@@ 20481 32768 |@5633 65536 | 1605 131072 | 1339 262144 | 957 524288 | 380 1048576 | 143 2097152 | 21 4194304 | 0 NFSv3 read/write by host (total us): x.16.0.x 3647019 x.16.0.x 8734488 x.16.0.x 50890034 x.16.0.x 68852624 x.16.0.x 70407241 x.16.0.x 79028592 x.16.0.x 83013688 x.16.0.x 104045281 x.16.0.x 105245138 x.16.0.x 124286383 x.16.0.x154526695 x.16.0.x 194419023 x.16.0.x 221794650 x.16.0.x 259302970 x.16.0.x 289694542 x.16.0.x 290207418 x.16.0.x 487983050 x.16.0.x1267135728 NFSv3 read/write top 10 files (total us): /teste/file1 95870303 /teste/file2 104212948 /teste/file3 104311607 /teste/file4 121076447 /teste/file5 137687236 /teste/file6 160895273 /teste/file7 180765880 /teste/file8 198827114 /teste/file9 372380414 /teste/file10 1126991407 -- cut here -- Max, will be difficult disable processors on that machine (production). Thanks again! Leal [http://www.eall.com.br/blog] -- This message posted fro
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Marcelo Leal wrote: >> Marcelo Leal wrote: >> >>> Hello all... >>> Thanks a lot for the answers! I think the problem >>> >> is "almost" fixed. Every dtrace documentation says to >> use predicates to guarantee the relation between the >> start/done probes... Max was the only one paying >> attention reading the docs. ;-) >> >>> But i'm still getting weird numbers: >>> >>> >> Which numbers don't look right? 3 of your reads took >> between 2 and 4 >> milliseconds, most were between >> 8 and 16 nanoseconds. 21 writes took between 2 and 4 >> milliseconds, the >> most amount of time >> spent doing read/write by host is about 1.2 seconds, >> and teste/file10 >> took about 1.1 seconds. >> Looks pretty good to me.(?). I'm curious about what >> you were expecting >> to see. >> > > The problem is the total numbers... > 1267135728 and 1126991407, for example. > 21 and 19 minutes in a ten minutes trace. > Or am i missing something? > When I do the arithmetic, I get about 1.2 seconds for the first number, and 1.1 seconds for the second number. These numbers are in nanoseconds, no? So, 1267135728/(1000*1000*1000) = 1.267... seconds. max > Leal > [http://www.eall.com.br/blog] > > >>> Wed Dec 10 08:36:33 BRST 2008 >>> Wed Dec 10 08:46:55 BRST 2008 >>> >>> cut here - >>> Tracing... Hit Ctrl-C to end. >>> ^C >>> NFSv3 read/write distributions (us): >>> >>> read >>>value - Distribution >>> >> - count >> >>>2 | >>> >> 0 >> 631 >> @@@ 145603 >> >>> 16 |@ >>> >>155926 >> 15970 >>6111 >> 942 >>372 >> 883 >>1649 >> 1090 >>8278 >> 24605 >>8868 >> 1694 >>304 >> 63 >>27 >> 31 >>43 >> 3 >>0 >> value - Distribution - >> count >> 128 | >> 0 >> 1083 >> @@@ 32622 >> >>> 1024 |@ >>> >>70353 >> 70851 >>47906 >> 44898 >>20481 >> 5633 >>1605 >> 1339 >>957 >> 380 >>143 >> 21 >>0 >> otal us): >> >>> x.16.0.x >>> >> 647019 >> >>> x.16.0.x >>> >> 734488 >> >>> x.16.0.x >>> >> 0890034 >> >>> x.16.0.x >>> >> 8852624 >> >>> x.16.0.x >>> >> 0407241 >> >>> x.16.0.x >>> >> 9028592 >> >>> x.16.0.x >>> >> 3013688 >> >>> x.16.0.x >>> >> 04045281 >> >>> x.16.0.x >>> >> 05245138 >> >>> x.16.0.x >>> >> 24286383 >> >>> x.16.0.x >>> >> 54526695 >> >>> x.16.0.x >>> >> 94419023 >> >>> x.16.0.x >>> >> 21794650 >> >>> x.16.0.x >>> >> 59302970 >> >>> x.16.0.x >>> >> 89694542 >> >>> x.16.0.x >>> >> 90207418 >> >>> x.16.0.x >>> >> 87983050 >> >>> x.16.0.x >>> >> 267135728 >> >>> NFSv3 read/write top 10 files (total us): >>> >>> /teste/file1 95870303 >>> /teste/file2 104212948 >>> /teste/file3 104311607 >>> /teste/file4 121076447 >>> /teste/file5 137687236 >>> /teste/file6 160895273 >>> /teste/file7 180765880 >>> /teste/file8 198827114 >>> /teste/file9 372380414 >>> /teste/file10 1126991407 >>> -- cut here -- >>> >>> >>> Max, will be difficult disable processors on that >>> >> machine (production). >> >>> >>> >> Yes. I understand. >> Regards, >> max >> >> >>> Thanks again! >>> >>> Leal >>> [http://www.eall.com.br/blog] >>> >>> >> ___ >> dtrace-discuss mailing l
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
> Marcelo Leal wrote: > > Hello all... > > Thanks a lot for the answers! I think the problem > is "almost" fixed. Every dtrace documentation says to > use predicates to guarantee the relation between the > start/done probes... Max was the only one paying > attention reading the docs. ;-) > > > Actually, this is not from reading the docs. It is > from being burnt a > few times by > getting "impossible" time values. By the way, I am > replying to you and > to the mailing > list, but messages to you are getting bounced. Oh, seems like i need to update my profile. ;-) > > But i'm still getting weird numbers: > > > Which numbers don't look right? 3 of your reads took > between 2 and 4 > milliseconds, most were between > 8 and 16 nanoseconds. 21 writes took between 2 and 4 > milliseconds, the > most amount of time > spent doing read/write by host is about 1.2 seconds, > and teste/file10 > took about 1.1 seconds. > Looks pretty good to me.(?). I'm curious about what > you were expecting > to see. The problem is the total numbers... 1267135728 and 1126991407, for example. 21 and 19 minutes in a ten minutes trace. Or am i missing something? Leal [http://www.eall.com.br/blog] > > > Wed Dec 10 08:36:33 BRST 2008 > > Wed Dec 10 08:46:55 BRST 2008 > > > > cut here - > > Tracing... Hit Ctrl-C to end. > > ^C > > NFSv3 read/write distributions (us): > > > > read > >value - Distribution > - count > >2 | > 0 > 631 > @@@ 145603 > > 16 |@ >155926 > 15970 >6111 > 942 >372 > 883 >1649 > 1090 >8278 > 24605 >8868 > 1694 >304 > 63 >27 > 31 >43 > 3 >0 > value - Distribution - > count > 128 | > 0 > 1083 > @@@ 32622 > > 1024 |@ >70353 > 70851 >47906 > 44898 >20481 > 5633 >1605 > 1339 >957 > 380 >143 > 21 >0 > otal us): > > > > x.16.0.x > > 647019 > > x.16.0.x > > 734488 > > x.16.0.x > > 0890034 > > x.16.0.x > > 8852624 > > x.16.0.x > > 0407241 > > x.16.0.x > > 9028592 > > x.16.0.x > > 3013688 > > x.16.0.x > > 04045281 > > x.16.0.x > > 05245138 > > x.16.0.x > > 24286383 > > x.16.0.x > > 54526695 > > x.16.0.x > > 94419023 > > x.16.0.x > > 21794650 > > x.16.0.x > > 59302970 > > x.16.0.x > > 89694542 > > x.16.0.x > > 90207418 > > x.16.0.x > > 87983050 > > x.16.0.x > > 267135728 > > > > NFSv3 read/write top 10 files (total us): > > > > /teste/file1 95870303 > > /teste/file2 104212948 > > /teste/file3 104311607 > > /teste/file4 121076447 > > /teste/file5 137687236 > > /teste/file6 160895273 > > /teste/file7 180765880 > > /teste/file8 198827114 > > /teste/file9 372380414 > > /teste/file10 1126991407 > > -- cut here -- > > > > > Max, will be difficult disable processors on that > machine (production). > > > Yes. I understand. > Regards, > max > > > Thanks again! > > > > Leal > > [http://www.eall.com.br/blog] > > > > ___ > dtrace-discuss mailing list > dtrace-discuss@opensolaris.org -- This message posted from opensolaris.org ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Marcelo Leal wrote: > Hello all... > Thanks a lot for the answers! I think the problem is "almost" fixed. Every > dtrace documentation says to use predicates to guarantee the relation between > the start/done probes... Max was the only one paying attention reading the > docs. ;-) > Actually, this is not from reading the docs. It is from being burnt a few times by getting "impossible" time values. By the way, I am replying to you and to the mailing list, but messages to you are getting bounced. > But i'm still getting weird numbers: > Which numbers don't look right? 3 of your reads took between 2 and 4 milliseconds, most were between 8 and 16 nanoseconds. 21 writes took between 2 and 4 milliseconds, the most amount of time spent doing read/write by host is about 1.2 seconds, and teste/file10 took about 1.1 seconds. Looks pretty good to me.(?). I'm curious about what you were expecting to see. > Wed Dec 10 08:36:33 BRST 2008 > Wed Dec 10 08:46:55 BRST 2008 > > cut here - > Tracing... Hit Ctrl-C to end. > ^C > NFSv3 read/write distributions (us): > > read >value - Distribution - count >2 | 0 >4 | 631 >8 | 145603 > 16 |@155926 > 32 |@@ 15970 > 64 |@6111 > 128 | 942 > 256 | 372 > 512 | 883 > 1024 | 1649 > 2048 | 1090 > 4096 |@8278 > 8192 |@@@ 24605 >16384 |@8868 >32768 | 1694 >65536 | 304 > 131072 | 63 > 262144 | 27 > 524288 | 31 > 1048576 | 43 > 2097152 | 3 > 4194304 | 0 > > write >value - Distribution - count > 128 | 0 > 256 | 1083 > 512 | 32622 > 1024 |@70353 > 2048 |@@ 70851 > 4096 |@@ 47906 > 8192 |@@ 44898 >16384 |@@@ 20481 >32768 |@5633 >65536 | 1605 > 131072 | 1339 > 262144 | 957 > 524288 | 380 > 1048576 | 143 > 2097152 | 21 > 4194304 | 0 > > NFSv3 read/write by host (total us): > > x.16.0.x 3647019 > x.16.0.x 8734488 > x.16.0.x 50890034 > x.16.0.x 68852624 > x.16.0.x 70407241 > x.16.0.x 79028592 > x.16.0.x 83013688 > x.16.0.x 104045281 > x.16.0.x 105245138 > x.16.0.x 124286383 > x.16.0.x154526695 > x.16.0.x 194419023 > x.16.0.x 221794650 > x.16.0.x 259302970 > x.16.0.x 289694542 > x.16.0.x 290207418 > x.16.0.x
Re: [dtrace-discuss] Round four: Re: code review req: 6750659 drti.ocrashes app due to corrupt environment
Mike Gerdts wrote:
>
> I believe that I have incorporated all of the feedback given
> (thanks!). Changes since the 2008-11-16 version include:
>
> - ksh style & coding standards compliance in test script (Roland)
> - "dof_init_debug == B_FALSE" vs. "!dof_init_debug" (Adam)
>
> The updated webrev is at:
>
> http://cr.opensolaris.org/~mgerdts/6750659-2008-12-03/
Quick review (patch code is quoted with "> "):
-- snip --
[snip]
> +
> +PATH=/usr/bin:/usr/sbin:$PATH
Is "export" not needed in this case, e.g. shouldn't subprocesses inherit
PATH, too ?
> +if (( $# != 1 )); then
> + print -u2 'expected one argument: '
> + exit 2
> +fi
> +
> +#
> +# jdtrace does not implement the -h option that is required to generate
> +# C header files.
> +#
> +if [[ "$1" == */jdtrace ]]; then
> + exit 0
Is the zero return code (= "suceess") Ok in this case ?
> +fi
> +
> +dtrace="$1"
> +startdir="$PWD"
> +dir=$(mktemp -td drtiXX)
> +if (( $? != 0 )); then
> + print -u2 'Could not create safe temporary directory'
> + exit 2
> +fi
> +
> +cd "$dir"
> +
> +cat > Makefile < +all: main
> +
> +main: main.o prov.o
> + \$(CC) -o main main.o prov.o
> +
> +main.o: main.c prov.h
> + \$(CC) -c main.c
> +
> +prov.h: prov.d
> + $dtrace -h -s prov.d
> +
> +prov.o: prov.d main.o
> + $dtrace -G -32 -s prov.d main.o
> +EOF
> +
> +cat > prov.d < +provider tester {
> + probe entry();
> +};
> +EOF
> +
> +cat > main.c < +#include
> +#include
> +#include "prov.h"
> +
> +int
> +main(int argc, char **argv, char **envp)
> +{
> + envp[0] = (char*)0xff;
> + TESTER_ENTRY();
> + return 0;
ISO C defines |EXIT_SUCCESS| (it's defined as |#define EXIT_SUCCESS
(0)|) ... I don't know whether it's better or worse in this case...
> +}
> +EOF
> +
> +make > /dev/null
> +status=$?
> +if (( $status != 0 )) ; then
Umpf... please remove the extra '$' - see
http://www.opensolaris.org/os/project/shell/shellstyle/#avoid_unneccesary_string_number_conversions
-- snip --
The remaining stuff looks good... :-)
Bye,
Roland
--
__ . . __
(o.\ \/ /.o) [EMAIL PROTECTED]
\__\/\/__/ MPEG specialist, C&&JAVA&&Sun&&Unix programmer
/O /==\ O\ TEL +49 641 3992797
(;O/ \/ \O;)
___
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
I think (us) is microseconds. There is one division by "1000" on the source code... Leal [http://www.eall.com.br/blog] -- This message posted from opensolaris.org ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Round four: Re: code review req: 6750659 drti.ocrashes app due to corrupt environment
I'd like to put an end to this discussion. Mike has done enough work,
and these tests are more than up to the task. If there are further
comments on the code, that's fine, but we can suffer through tests
that are functionally correct, but stylistically less than perfect.
Adam
On Dec 10, 2008, at 6:13 AM, Roland Mainz wrote:
> Mike Gerdts wrote:
>>
>> I believe that I have incorporated all of the feedback given
>> (thanks!). Changes since the 2008-11-16 version include:
>>
>> - ksh style & coding standards compliance in test script (Roland)
>> - "dof_init_debug == B_FALSE" vs. "!dof_init_debug" (Adam)
>>
>> The updated webrev is at:
>>
>> http://cr.opensolaris.org/~mgerdts/6750659-2008-12-03/
>
> Quick review (patch code is quoted with "> "):
> -- snip --
> [snip]
>> +
>> +PATH=/usr/bin:/usr/sbin:$PATH
>
> Is "export" not needed in this case, e.g. shouldn't subprocesses
> inherit
> PATH, too ?
>
>> +if (( $# != 1 )); then
>> + print -u2 'expected one argument: '
>> + exit 2
>> +fi
>> +
>> +#
>> +# jdtrace does not implement the -h option that is required to
>> generate
>> +# C header files.
>> +#
>> +if [[ "$1" == */jdtrace ]]; then
>> + exit 0
>
> Is the zero return code (= "suceess") Ok in this case ?
>
>> +fi
>> +
>> +dtrace="$1"
>> +startdir="$PWD"
>> +dir=$(mktemp -td drtiXX)
>> +if (( $? != 0 )); then
>> + print -u2 'Could not create safe temporary directory'
>> + exit 2
>> +fi
>> +
>> +cd "$dir"
>> +
>> +cat > Makefile <> +all: main
>> +
>> +main: main.o prov.o
>> + \$(CC) -o main main.o prov.o
>> +
>> +main.o: main.c prov.h
>> + \$(CC) -c main.c
>> +
>> +prov.h: prov.d
>> + $dtrace -h -s prov.d
>> +
>> +prov.o: prov.d main.o
>> + $dtrace -G -32 -s prov.d main.o
>> +EOF
>> +
>> +cat > prov.d <> +provider tester {
>> + probe entry();
>> +};
>> +EOF
>> +
>> +cat > main.c <> +#include
>> +#include
>> +#include "prov.h"
>> +
>> +int
>> +main(int argc, char **argv, char **envp)
>> +{
>> + envp[0] = (char*)0xff;
>> + TESTER_ENTRY();
>> + return 0;
>
> ISO C defines |EXIT_SUCCESS| (it's defined as |#define EXIT_SUCCESS
> (0)|) ... I don't know whether it's better or worse in this case...
>
>> +}
>> +EOF
>> +
>> +make > /dev/null
>> +status=$?
>> +if (( $status != 0 )) ; then
>
> Umpf... please remove the extra '$' - see
> http://www.opensolaris.org/os/project/shell/shellstyle/#avoid_unneccesary_string_number_conversions
> -- snip --
>
> The remaining stuff looks good... :-)
>
>
>
> Bye,
> Roland
>
> --
> __ . . __
> (o.\ \/ /.o) [EMAIL PROTECTED]
> \__\/\/__/ MPEG specialist, C&&JAVA&&Sun&&Unix programmer
> /O /==\ O\ TEL +49 641 3992797
> (;O/ \/ \O;)
> ___
> dtrace-discuss mailing list
> dtrace-discuss@opensolaris.org
--
Adam Leventhal, Fishworkshttp://blogs.sun.com/ahl
___
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Hi Marcelo, Marcelo Leal wrote: > I think (us) is microseconds. There is one division by "1000" on the source > code... > > Oops. You're right. I did not see that. (That might explain the 4-8 nanosecond I/Os, which I did think seemed pretty fast. They are actually 4-8 microsecond). So, you want to know how you can spend 19 or 20 minutes in a 10 minute trace? You have multiple cpu's, so each cpu can be working in parallel on different I/O requests. If you have 8 cpu's, the average time spent by each cpu would be about 2.5 minutes. This does sound a little high to me, but not extreme. If you got 80 minutes, I would be concerned that all cpus are working on requests all the time. It might be difficult to correlate per cpu, as I suspect the start and done probe for a given I/O could fire on different cpus. max > Leal > [http://www.eall.com.br/blog] > ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Ok, but that is a bug, or should work like that? We can not use dtrace on multiple processors systems? Sorry, but i don't get it... __ Leal [http://www.eall.com.br/blog] -- This message posted from opensolaris.org ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Hi Marcelo, Marcelo Leal wrote: > Ok, but that is a bug, or should work like that? > We can not use dtrace on multiple processors systems? > Sorry, but i don't get it... > I don't consider this a bug. I think it depends on what you are trying to measure. The script you are using measures latency for read/write operations across all cpus. There is nothing wrong with the sum of the times being longer than the total time you are tracing, if there are multiple cpus. If you wanted to measure latency per cpu, the script would need to be changed. So, what are you trying to measure? I think more interesting is to find out why some I/O operations take longer than others. For this, you can use speculative tracing. Trace all function entries and returns, but only commit if the time spent is longer than the longest time found thus far. I'm sure others have ideas about this... max > __ > Leal > [http://www.eall.com.br/blog] > ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Sorry, but i do not agree. We are talking about a NFSv3 provider, and not about how many cpu's there are on the system. I do not have the knowledge to discuss with you the aspects about the implementation, but as a user point of view, i think that numbers don't make sense. If the fact that the number of cpu's is important for the start/done for the NFS provider, i think it will for all other dtrace providers. Thanks a lot for your answers Max! __ Leal [http://www.eall.com.br/blog] -- This message posted from opensolaris.org ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
No bug here - we can absolutely use DTrace on MP systems, reliably and with confidence. The script output shows some nasty outliers for a small percentage of the reads and writes happening on the server. Time to take a closer look at the IO subsystem. I'd start with "iostat -znx 1", and see what the queues look like, IO rates, and service times. I'd also have a look at the network. Download nicstat and run it (go to blogs.sun.com and search for "nicstat" - it's easy to find). What are you using for an on-disk file system? UFS or ZFS? /jim Marcelo Leal wrote: > Ok, but that is a bug, or should work like that? > We can not use dtrace on multiple processors systems? > Sorry, but i don't get it... > __ > Leal > [http://www.eall.com.br/blog] > ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Printing the main script responsible for writing a file
On Mon, Nov 24, 2008 at 06:03:44AM -0800, debabrata das wrote:
> Hi all,
>
> I am new in dtrace. But I have urgent requirement. I would like to know a
> name of program which is responsible for updating a file. I have written the
> small following program :
>
> #!/usr/sbin/dtrace -s
>
> syscall::open*:entry
> /pid == $1/
> {
>printf("%s %s %d", execname, copyinstr(arg0), pid);
> }
>
> Now if I ruun this then I get following output :
>
> -bash-3.00$ ./text.d | grep deba
> dtrace: script './text.d' matched 2 probes
> 8402 open64:entry bash /tmp/deba.txt 18786
>
> Now I want to know which script is updating this file. I would like to the
> see the output of "ptree 18786". I am not sure how to this structure of ptree
> or the name of the script.
you want curpsinfo->pr_psargs; it's a string which contains the string that
"ptree" outputs, which is the "argv" array, separated by spaces, truncated
to 80 characters.
Cheers,
- jonathan
___
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
[dtrace-discuss] How to tell FPU activity on a CMT processor? Which tool, what to look at?
Okay, I admit to confusion and a certain amount of defeat. I have numerous T2000 boxes and I want to see how much floating point activity is taking place on them in order to tell if I should move applications to a T2 based box. So I go looking around. First up is the cooltst application from http://cooltools.sunsource.net/cooltst/ which I try to download. All I can get is a Zip file instead of the cooltst_v3.tar.bz2 file mentioned in the download section of the page. So the question here is how/where to get a version of cooltst that runs on a T2000? Next up is other mechanisms to get FPU info. So I go searching the web and find Darryl Gove's web page of http://blogs.sun.com/d/entry/using_dtrace_to_locate_floating that talks to using kstat and dtrace to look for 'fpu_unfinished_traps'. So I use both kstat and dtrace options that Darryl mentions and both options come up with zero (0) counts! So I start wondering about the count actually being zero and go looking some more. Then I find, http://www.informit.com/articles/article.aspx?p=1161980&seqNum=3 and read through section 4.3.8 where is mentions "One of the metrics that kstat reports is the number of emulated floating-point instructions. Not all floating-point operations are performed in hardware; some have been left to software.". So now I am wondering if there is any pure hardware based FPU transactions taking place that would NOT show up in the 'fpu_unfinished_traps' counts on a CMT T1 chip based server? And if so, how do I find out about those operations? So, the question is, lacking cooltst that I can get which is usable on a T2000, what is the full story on floating point math on a CMT and where/how can data about it's usage be gotten? My thanks for any and all help in this! pdc -- This message posted from opensolaris.org ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
[dtrace-discuss] dtrace (1M) manpage/ docs seem incorrect re 32/64 bit
The dtrace(1M) manpage seems to be incorrect, it states: -32 | -64 The D compiler produces programs using the native data model of the operating system kernel. You can use the isainfo -b command to determine the current operating system data model. If the -32 option is specified, dtrace forces the D compiler to compile a D program using the 32-bit data model. If the -64 option is speci- fied, dtrace forces the D compiler to compile a D pro- gram using the 64-bit data model. These options are typ- ically not required as dtrace selects the native data model as the default. The data model affects the sizes of integer types and other language properties. D pro- grams compiled for either data model can be executed on both 32-bit and 64-bit kernels. The -32 and -64 options also determine the ELF file format (ELF32 or ELF64) pro- duced by the -G option. If I read this correctly then trying doing "dtrace -G -o foo ..." will default to being 32-bit on a 32-bit system and 64-bit on a 64-bit system. That does not seem to be the case however - it always seems to do a 32-bit build if -64 is not specified. Further, on systems where bug 6456626 has been fixed, -G will try to guess whether to do 32-bit or 64-bit compile and so the docs need to be updated to reflect this as well. Cheers, David Holmes -- This message posted from opensolaris.org ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] How to tell FPU activity on a CMT processor? Which tool, what to look at?
On 12/11/08, Paul Clayton <[EMAIL PROTECTED]> wrote: > Next up is other mechanisms to get FPU info. > > So I go searching the web and find Darryl Gove's web page of > http://blogs.sun.com/d/entry/using_dtrace_to_locate_floating that talks to > using kstat and dtrace to look for 'fpu_unfinished_traps'. So I use both > kstat and dtrace options that Darryl mentions and both options come up with > zero (0) counts! Take a look at this doc: http://www.sun.com/blueprints/1205/819-5144.pdf > So now I am wondering if there is any pure hardware based FPU transactions > taking place that would NOT show up in the 'fpu_unfinished_traps' counts on a > CMT T1 chip based server? And if so, how do I find out about those operations? You can read the full micro architecture of the T1, specifically "OpenSPARC T1 Micro Architecture Specification", chapter 7: Floating-Point Unit. http://www.opensparc.net/opensparc-t1/index.html Rayson > > So, the question is, lacking cooltst that I can get which is usable on a > T2000, what is the full story on floating point math on a CMT and where/how > can data about it's usage be gotten? > > My thanks for any and all help in this! > > pdc > -- > This message posted from opensolaris.org > ___ > dtrace-discuss mailing list > dtrace-discuss@opensolaris.org > ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] How to tell FPU activity on a CMT processor? Which tool, what to look at?
Rayson.. My thanks for the pointers to the two documents. Very interesting reading. For the emulated instructions, I now have my answers to tell what is happening. In the case currently under investigation the only emulation is FSQRTD and that stands at 884K but is not currently growing so some other app then the one in question use FSQRTD. But in reading the T1 specification, there appears to be a lot of FPU 'stuff' that takes place within the single instance FPU with no emulation. So part of my question remains. How can I find out about FPU usage for non-emulated instructions on a T1 chipset? I have to wonder about the command "/usr/sbin/cpustat -c pic0=FP_instr_cnt,pic1=Instr_cnt,sys 1 10" and if this would deal with hardware based FPU usage or is another option for the emulated functions? The document does not provide a clear hint on this subject. Take care. pdc -Original Message- From: Rayson Ho [mailto:[EMAIL PROTECTED] Sent: Thursday, December 11, 2008 1:12 AM To: Clayton, Paul D Cc: dtrace-discuss@opensolaris.org Subject: Re: [dtrace-discuss] How to tell FPU activity on a CMT processor? Which tool, what to look at? On 12/11/08, Paul Clayton <[EMAIL PROTECTED]> wrote: > Next up is other mechanisms to get FPU info. > > So I go searching the web and find Darryl Gove's web page of http://blogs.sun.com/d/entry/using_dtrace_to_locate_floating that talks to using kstat and dtrace to look for 'fpu_unfinished_traps'. So I use both kstat and dtrace options that Darryl mentions and both options come up with zero (0) counts! Take a look at this doc: http://www.sun.com/blueprints/1205/819-5144.pdf > So now I am wondering if there is any pure hardware based FPU transactions taking place that would NOT show up in the 'fpu_unfinished_traps' counts on a CMT T1 chip based server? And if so, how do I find out about those operations? You can read the full micro architecture of the T1, specifically "OpenSPARC T1 Micro Architecture Specification", chapter 7: Floating-Point Unit. http://www.opensparc.net/opensparc-t1/index.html Rayson > > So, the question is, lacking cooltst that I can get which is usable on a T2000, what is the full story on floating point math on a CMT and where/how can data about it's usage be gotten? > > My thanks for any and all help in this! > > pdc > -- > This message posted from opensolaris.org > ___ > dtrace-discuss mailing list > dtrace-discuss@opensolaris.org > ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
Re: [dtrace-discuss] Is the nfs dtrace script right (from nfsv3 provider wiki)?
Hi Marcelo, Marcelo Leal wrote: > Sorry, but i do not agree. > We are talking about a NFSv3 provider, and not about how many cpu's there > are on the system. I do not have the knowledge to discuss with you the > aspects about the implementation, but as a user point of view, i think that > numbers don't make sense. If the fact that the number of cpu's is important > for the start/done for the NFS provider, i think it will for all other dtrace > providers. > I have found that almost all "bugs" with dtrace are either in the scripting or in the interpretation of the output. The mechanism used to implement dtrace is quite simple, and for the script you are running, I don't believe you are hitting any bug in the implementation of dtrace. Also, using dtrace to examine the system is like using a microscope. You need to know the big picture first, before you can determine what you need to trace. Otherwise you can end up getting a lot of detail about something that has nothing to do with the problem you are experiencing. In this instance, as Jim says, iostat will give you a better view of the big picture. As for the 20 minutes total time spent for I/O in a script running 10 minutes, this could happen even on a single CPU. For example, you start the script and immediately 10 I/O requests are made. Now, let's say all 10 requests are queued (i.e., block, waiting for some resource). If they all finish just before the end of the 10 minute period, the total elapsed time would be about 100 minutes. So long as the total time divided by the number of operations does not exceed 10 minutes, the output is reasonable. I suspect the outliers, (the I/Os that are taking 2-4 seconds) are due to queuing at the disk driver, but they could be due to scheduling as well. (I would have to look at exactly when the start and done probes fire to determine this. But fortunately, source code is available to determine this.). As I mentioned in an earlier post, speculative tracing can be used to determine the code path taken for the longest I/O. I have written a script that might work, but have no way to test it at this time. If you are interested, I'll post it. Jim, as for taking this offline, that is ok for you, but my posts to Marcelo are bouncing... max > Thanks a lot for your answers Max! > > __ > Leal > [http://www.eall.com.br/blog] > ___ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
