LHC@home

Thank Markku for enable this THING it helps a lot to control the Work we have do
rutsch mer doch dr Buckel na schofseggel
jetzt sind es sogar Pending credit: 537.01

Pending credit: 2155.83

2021.41 and more to come.

OHH NO

Pending credit: 5143.44

no comment.

I don't know, if it's really better to know or not..



AUF DIE DAUER HILFT NUR POWER

> Pending credit: 5143.44
>
> no comment.
>
> I don't know, if it's really better to know or not..
>
>
>
> AUF DIE DAUER HILFT NUR POWER
>

I have only 1000 Pending credit, but that's from one machine only.. :-/

sat 10.22.2004 -11 am
are 613
rutsch mer doch dr Buckel na -the jokes must go on

Close to 3000 pending credits as of this morning, don't know whats going on with the validator. I seem to be getting Credits but the pending have doubled in the last 2 days ...

> Close to 3000 pending credits as of this morning, don't know whats going on
> with the validator. I seem to be getting Credits but the pending have doubled
> in the last 2 days ...

Most of the work units you have pending are waiting for enough matching results. Pentiums and Athlons are again giving different results. So it might take some time but ultimately you should get some credit.

3 identical results are required, and when a new workunit is made, 4 results are sent out. If they don't return as identical, one more result is sent out, which may take a few days (or even two weeks) to complete. If a quorum can't be reached even then, a new result is generated and so on until quorum of 3 identical results is achieved or maximum number of successful results (10) is reached.

Markku Degerholm
LHC@home Admin

Thats what I figured Markku, I wasn't that worried about it though. Plus I've been slowly adding more PC's & Faster PC's to the crunching so thats probably got something to do with it to... :)

>> Pentiums and Athlons are again giving different results.

Isn't this an argument for splitting Pentium and Athlon systems into two different platforms and using homogeneous redundancy within platforms?


Giskard - the first telepathic robot.

> >> Pentiums and Athlons are again giving different results.
>
> Isn't this an argument for splitting Pentium and Athlon systems into two
> different platforms and using homogeneous redundancy within platforms?

To my knowledge, boinc doesn't support this kind of platform splitting. If we enabled homogenous redundancy, we would get about 500 different platforms (processors with different CPU clocks would be classified as different platforms).
I'm afraid this would make work unit finishing a lot longer process because it is required that there are enough hosts per each platform.

Another thing is that we are still trying to find out how to make Pentiums and Athlons produce identical results. It should be possible, after all.

Markku Degerholm
LHC@home Admin

> To my knowledge, boinc doesn't support this kind of platform splitting. If we
> enabled homogenous redundancy, we would get about 500 different platforms
> (processors with different CPU clocks would be classified as different
> platforms).

This excerpt on homogeneous redundancy taken from the BOINC pages here

-------------------------------------------------------------------------
When this feature is enabled, the BOINC scheduler will send results for a given workunit only to hosts with the same operation system name and CPU vendor (i.e., the os_name and p_vendor fields of the host description). For example: if a result has been sent to a host of type (Windows XP, Intel), then other results of that workunit will only be sent to hosts of type (Windows XP, Intel).
---------------------------------------------------------------------------

This implies that it should be possible to distinguish between Intel and AMD processors without regard to other processor parameters. It might be necessary to establish different set-ups for different versions of Windows (98, 2000, XP, etc), but even that might possibly be simplified.

I'd guess Rom Walton at Berkeley is the guy who'd know.

Edit: I just checked the scheduler source code. The homogeneous redundancy feature checks the processor type for the strings 'Intel', 'AMD' and 'Macintosh'. OS names checked are 'Windows', 'Linux', 'Darwin' and 'SunOS'. It seems likely that just turning on homogeneous redundancy will eliminate the sort of cross-platform inconsistencies you're getting at a stroke, and without setting up any new platforms.



Giskard - the first telepathic robot.

>
> To my knowledge, boinc doesn't support this kind of platform splitting. If we
> enabled homogenous redundancy, we would get about 500 different platforms
> (processors with different CPU clocks would be classified as different
> platforms).

Taking a quick look on checkin_notes and accompanying code-changes, this requirement was loosened 17. June and optimized a little later so looks like this:
http://boinc.berkeley.edu/cgi-bin/cvsweb.cgi/boinc/sched/sched_send.C.diff?r1=1.25&r2=1.26&f=h

Looking at this code, for LHC homogenous redundancy only checks if Intel, AMD or unknown, and Windows, linux or unknown. It doesn't even differensiate between win9x & NT, so it shouldn't be any problem enabling homogenous redundancy.

> I'm afraid this would make work unit finishing a lot longer process because it
> is required that there are enough hosts per each platform.
>

In LHC it would worse-case be 9 different, if someone have non-detected OS...
Maybe it will take longer time, but for most less re-issuing of wu to get a quorum will probably speed things up instead of slowing it down.

Pending credit: 8619.09
03:06 31.10.2004

Irgend etwas klemmt f�rcherlich oder ist zu langsam in der ganzen Verarbeitungskette

Something is hanging in a ugly way or something is to slow in processing chain

edit:
Pending credit: 8775.42

Markku,

Could you expand on your statement "Pentiums and Athlons are again giving different results"? How do the results differ? I'm not trying to start an AMD vs. Intel war, I'm just curious about the differences. Was there a time the two processors were giving the same or almost the same results (within whatever difference range you deem acceptable)?

Ingleside,

Looks to me as though it checks for a Macintosh host and Darwin and SunOS OS's, too. I believe that would make 20 different platforms, not nine.

> Looking at this code, for LHC homogenous redundancy only checks if Intel, AMD or unknown,
> and Windows, linux or unknown. It doesn't even differensiate between win9x & NT, so it
> shouldn't be any problem enabling homogenous redundancy.

I believe that would make 20 different platforms, not nine.

> In LHC it would worse-case be 9 different,

From the referenced "Diff for /boinc/sched/sched_send.C between version 1.25 and 1.26"
http://boinc.berkeley.edu/cgi-bin/cvsweb.cgi/boinc/sched/sched_send.C.diff?r1=1.25&r2=1.26&f=h

int OS(SCHEDULER_REQUEST& sreq){
if ( strstr(sreq.host.os_name, "Linux") != NULL ) return Linux;
else if( strstr(sreq.host.os_name, "Windows") != NULL ) return Windows;
else if( strstr(sreq.host.os_name, "Darwin") != NULL ) return Darwin;
else if( strstr(sreq.host.os_name, "SunOS") != NULL ) return SunOS;
else return noos;
};

int CPU(SCHEDULER_REQUEST& sreq){
if ( strstr(sreq.host.p_vendor, "Intel") != NULL ) return Intel;
else if( strstr(sreq.host.p_vendor, "AMD") != NULL ) return AMD;
else if( strstr(sreq.host.p_vendor, "Macintosh") != NULL ) return Macintosh;
else return nocpu;
};

Cheers,

Stephen
<a> [/url]

> I believe that would make 20 different platforms, not nine.
>
> > In LHC it would worse-case be 9 different,
>

Actually, Stephen, the code imples NO additional platforms. The homogeneity information on processor and OS is extracted from data returned from each computer, and is processed independently of other platform consideration. Just turning on the feature will separate Pentiums from Athlons with no further input required!


Giskard - the first telepathic robot.

> Could you expand on your statement "Pentiums and Athlons are again giving
> different results"? How do the results differ? I'm not trying to start an AMD
> vs. Intel war, I'm just curious about the differences. Was there a time the
> two processors were giving the same or almost the same results (within
> whatever difference range you deem acceptable)?

The differences come mostly from EXP/LOG instructions, which have different precisions on AMD and Intel. Differences starts as small but cumulate during calculations. Some types of studies which do not use these instructions have much less differences. For differences in those studies we blame overclocked computers, memory errors and such, at least until proved otherwise.

> > Looking at this code, for LHC homogenous redundancy only checks if Intel,
> AMD or unknown,
> > and Windows, linux or unknown. It doesn't even differensiate between
> win9x & NT, so it
> > shouldn't be any problem enabling homogenous redundancy.
>
> I believe that would make 20 different platforms, not nine.
>
> > In LHC it would worse-case be 9 different,
>

I also recheched the BOINC code and revoke my previous statement of over 500 platforms. I think there would be 9 (3x3) platforms. But we don't want to enable homogeneous redundancy yet... We want identical results from every platform, if possible.

Markku Degerholm
LHC@home Admin

The Workunit 354189 has been reported succesfully by three different users and all three have pending credit with averages higher than 85.00 credits, however the credit has not been granted for several days now, are we waiting for some more results on the same unit?

I thought I had read three succesfull in total!


Patience is a virtue

> The Workunit 354189 has been reported succesfully by three different users and
> all three have pending credit with averages higher than 85.00 credits, however
> the credit has not been granted for several days now, are we waiting for some
> more results on the same unit?
>
> I thought I had read three succesfull in total!


I believe that the desired outcome is identical not just successful.

"We want identical results from every platform, if possible." Markku

>> I believe that the desired outcome is identical not just
> successful.
>
> "We want identical results from every platform, if possible."
> Markku

Well, what is identical on the units that have been granted credit is the credit requested by each host, but not necessarily the time that the CPU took in seconds to analize, I have several (almost all of the granted credit units) that show totally different processing times but the claimed credit is within a 0.01 to 0.02 plus or minus from what other users claimed.

If that is what you mean by "identical results"... then ok.

Thanks for the info.
Patience is a virtue