Advanced search

Forums : Technical Support : Is HR turned on?
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Conan
Avatar

Send message
Joined: 28 Aug 07
Posts: 169
Credit: 1,280,875
RAC: 0
Message 7259 - Posted: 13 Sep 2008, 7:26:36 UTC

Just wondering if HR is turned on as I seem to be lumped in with Intel when I am AMD.

This occurs on both Windows and Linux.

I have just been granted the pleasure of working on 39 Windows work units and 3 Linux work units.
Unfortunately nearly all of my wingmen are Intel.

Is this going to be a problem?
The OS seems to be ok as per HR (no Linux grouped with Windows).
ID: 7259 · Report as offensive     Reply Quote
Phoneman1

Send message
Joined: 5 Nov 07
Posts: 113
Credit: 3,100,327
RAC: 0
Message 7261 - Posted: 13 Sep 2008, 10:51:02 UTC - in response to Message 7259.  


Unfortunately nearly all of my wingmen are Intel.

Is this going to be a problem?
The OS seems to be ok as per HR (no Linux grouped with Windows).


September 5, 2008 Another note about test work
It should also be noted that test work will be validated regardless of whether or not it matches the partner WU. This will insure that nobody is penalized for running test work with the experimental HR.


The HR is meant to be coarser grained we were told somewhere. The same happened in last weekend\'s test run and I\'ve not seen a problem yet. All of my pending credits from last week have either not yet beem crunched by Intel or AMD wingmen but there is still plenty of time left for them to be returned. I do have the odd pending unit where no wingman has been assigned - just like some from the main pool of units.

As I understand things AMD and Intel processors can give slightly different answers in floating point calculations. The differences are minute in each such calculation and effect the least siugnificant digit. Depending on the program these can become more significant depending upon the operations (multiply and divide) performed and the order in which they are performed. If the program perfoms these operations and then rounds the answers off to a sufficently high level then it is possible that both Intel and AMD machines will return the same answer. Another option is to use a \"fudge factor\" in the validator (e.g. your answer is within 1% of the wingman\'s answer). Of course the project could use both options. I don\'t know which option(s) is / are being used here. I guess this is one of the main tests the project wand to carry out with these runs.

The business about keeping Linux and Windows separate is at first glance a little strange. How can an operating system influence a program running independently under it? The answer lies in code shared with the operating system. Code in DLL files can be used by both programs and operating systems. In one case you have one of the worlds\' biggest corporations pushing its latest fixes for operating systems and DLLs to its licenced users every second Tuesday in the month, unless those users take action on their machines to stop it.

In the other case you haven\'t. As a result the DLLs used by Linux users depend on which flavour of Linux they run, when they last updated their system etc, etc.

Incidentally, I get the impression that this is the reason why it seems that Linux users are usually the ones who see \"too many successful result\" messages. The answers from the wingmen aren\'t matching - presumably because of the variation in the DLLs being used. It might be interesting to investigate any Windows examples of this message. It could be one of the wingmen are running an old version of Windows.... Anyone seen anything like this?

Some projects seem to get over this problem by not using OS DLLs and providing all the code themselves - wasteful in program download times and maintenance for the project but it saves this particular problem, I guess.

Phoneman1
P.S. Conan, please turn the wick up that Windows AMD machine of yours; Windows Intel user here waiting patiently for his credits!!!
ID: 7261 · Report as offensive     Reply Quote
Profile Conan
Avatar

Send message
Joined: 28 Aug 07
Posts: 169
Credit: 1,280,875
RAC: 0
Message 7293 - Posted: 17 Sep 2008, 12:02:02 UTC - in response to Message 7261.  


Unfortunately nearly all of my wingmen are Intel.

Is this going to be a problem?
The OS seems to be ok as per HR (no Linux grouped with Windows).


September 5, 2008 Another note about test work
It should also be noted that test work will be validated regardless of whether or not it matches the partner WU. This will insure that nobody is penalized for running test work with the experimental HR.


The HR is meant to be coarser grained we were told somewhere. The same happened in last weekend\'s test run and I\'ve not seen a problem yet. All of my pending credits from last week have either not yet beem crunched by Intel or AMD wingmen but there is still plenty of time left for them to be returned. I do have the odd pending unit where no wingman has been assigned - just like some from the main pool of units.

As I understand things AMD and Intel processors can give slightly different answers in floating point calculations. The differences are minute in each such calculation and effect the least siugnificant digit. Depending on the program these can become more significant depending upon the operations (multiply and divide) performed and the order in which they are performed. If the program perfoms these operations and then rounds the answers off to a sufficently high level then it is possible that both Intel and AMD machines will return the same answer. Another option is to use a \"fudge factor\" in the validator (e.g. your answer is within 1% of the wingman\'s answer). Of course the project could use both options. I don\'t know which option(s) is / are being used here. I guess this is one of the main tests the project wand to carry out with these runs.

The business about keeping Linux and Windows separate is at first glance a little strange. How can an operating system influence a program running independently under it? The answer lies in code shared with the operating system. Code in DLL files can be used by both programs and operating systems. In one case you have one of the worlds\' biggest corporations pushing its latest fixes for operating systems and DLLs to its licenced users every second Tuesday in the month, unless those users take action on their machines to stop it.

In the other case you haven\'t. As a result the DLLs used by Linux users depend on which flavour of Linux they run, when they last updated their system etc, etc.

Incidentally, I get the impression that this is the reason why it seems that Linux users are usually the ones who see \"too many successful result\" messages. The answers from the wingmen aren\'t matching - presumably because of the variation in the DLLs being used. It might be interesting to investigate any Windows examples of this message. It could be one of the wingmen are running an old version of Windows.... Anyone seen anything like this?

Some projects seem to get over this problem by not using OS DLLs and providing all the code themselves - wasteful in program download times and maintenance for the project but it saves this particular problem, I guess.

Phoneman1
P.S. Conan, please turn the wick up that Windows AMD machine of yours; Windows Intel user here waiting patiently for his credits!!!


G\'Day Phoneman1,
It looks like my computer has heard your request and is currently going flat strap on Cosmology (High Priority) due to it downloading around 90 work units and then realising that they will take a lot longer than Boinc Manager thought.

Most are taking 5 or more hours and as this will require 5 days solid processing of Cosmo only, so my other two projects are doing nothing until all of Cosmo gets cleared out.

So you will get your credits.
ID: 7293 · Report as offensive     Reply Quote

Forums : Technical Support : Is HR turned on?