On my MacBookPro with Dual Core, my credits given over 10 days dropped from 17,000 running 2 FAMOUS tasks and since running 2 HadSM3 tasks, has dropped to 8,400 over 10 days.

Both were steadily keeping at these rates on the Statistics graph page, so, why? (I do not take any notice of so called recent average credits, because they have been altered beyond deserving any definition of "average".)

I do remember there was a problem some time ago with tasks competing for use of the CPU and slowing each other in the process of doing so. Could this be what is happening here? How shall I overcome this, as I am only being allocated HadSM3 tasks now?

Keith

There were problems with the compiler used with the FAMOUS models - it wasn't possible to turn off some of the optimisations that made the models more stable for Windows and Linux. As a result, the models run a lot faster on Mac computers, at the expense of greater stability.
This meant that they got more credits for a given time period than for the other 2 OS types, and for the FAMOUS models, the credits were increased to give parity for those computers. So you were getting a bonus.

People with all model types selected will get whatever is next in the queue at the time, which was apparently slab models at the time your computer requested replacements.
If credits are important to you, then just set your preferences to only get FAMOUS models.

---

Backups: Here

I do remember there was a problem some time ago with tasks competing for use of the CPU and slowing each other in the process of doing so. Could this be what is happening here? How shall I overcome this, as I am only being allocated HadSM3 tasks now?

Keith[/quote]

BTW, the models that slow each other down when run together are the Hadam3p’s. This is apparently caused because they are so memory intensive that they fight each other for access to the RAM. I have not herd of HadSM3 models, which use much less memory, acting this way.

---

Thanks for your reply Les.
I could not care less about the credits, apart from using them as a guide to see how my computer is working.
It is extraordinary that the FAMOUS tasks were clocking up double the rate credits "earned" of the HadSM3 tasks. And as I understand you, my MacBookPro was still doing the same amount of useful work on FAMOUS tasks, but was doubling the apparent worth according to the credits given. And I only noticed the drop at the end of the FAMOUS series of tasks, but not the increase on it commencing.

Keith

And thanks, Jim for putting me straight on my vague memory of the tasks fighting for the CPU.
Keith

When we only ran HadSM, HadSM-MH and HadCM models they were all given very similar rates of credit but I think the rates were always based on performance on the Windows compilation because that's what the majority of members use. I think Geophi said HadCM, for example, was slower on Linux.

When HadAM3P was launched it had far bigger uploads and downloads than anything we'd had before. The credit rate was increased to 'compensate' for this. The HadAM3P regional models are the same type so were given the same credits as the earlier HadAM3P. FAMOUS credits were set to be on a par with HadAM3P because the uploads and downloads are also pretty big. So really it's HadSM that's now out of line credit-wise.

Personally I think longer models like HadSM should get a credit bonus at the end to reward people for finishing them. But nobody else has ever expressed an opinion on this...

---

Cpdn news

mo.v

A bonus is one thing, but I have found on my MacBook that the difference is by a factor of TWO (i.e. DOUBLED). That is much more than a bonus. In any case, completing with FAMOUS was not an option of the cruncher, as the majority of the tasks failed long before completion anyway and the trickles still gave the same rate of credit on failed tasks.

So, I am perplexed by that reasoning.
As I have said, I am not complaining that I don't get credits. All I would like to see is a yardstick by which I can gauge what is happening. And FAMOUS should certainly not get DOUBLE the rate of credits compared to HadSM3, surely not. How can we know what bias there is attached to each model if we want to be happy with our computer is functioning correctly? Is there a list opf the "bias" attached to each model running? It seems pointless (Sorry for the pun!!) to have differing values for each model to attract crunchers to particular types of model if the bias is not made apparent.

It would otherwise make me think that the "credit system" is sinking into disrepute just has the so-called "recent average", which has long become meaningless as a yardstick since it was "sexed up" by the programmers.

Keith

I've already told you why Mac computers get more credits per "longish" unit of time (e.g 1 week), compared to running slabs.
(And why FAMOUS models on a Mac were more prone to failure.)

Get a Windows or Linux computer and try both model types on them, and you'll see that the credits are much the same for slabs and FAMOUS on either OS. This is because (again!), the FAMOUS models don't run so rapidly on those 2 OSs.

The speed of Macs was commented on several times on the beta site when we tested the FAMOUS models.

The credits for HadSM are probably about 15% lower than for HadAM3P (regional) and FAMOUS as explained above. FAMOUS could well be 30 or 40% faster on Mac than on Linux or Windows. If HadSM is slower on Mac than on Windows, that will pretty much explain the credit discrepancy you're seeing. As I said before, the rates are all fixed for Windows.

The compilations for apps are not designed to produce the same credit rate per crunching hour for all OSs running it. They're designed to make the app as stable as possible on each OS. As Les explained above, this wasn't fully achieved for Mac and FAMOUS. This app runs very fast but less stably than on Windows and Linux.

Credit parity between apps per crunching hour is calculated for Windows. This is done on the Beta project before the public launch. If apps are slow on Mac or Linux compared with Windows Tolu always tried to tweak the compilations to achieve greater parity between OSs but, as I said before, operating stability (valid results that the researchers like and model completion without crashes) is always given priority over crunching speed if there's a conflict.

The credits per completed model of every type, past and present, are listed in the News Archive on the independent forum in the post dated 9 Sep 2009 here:

http://climateprediction.net/board/viewtopic.php?f=36&t=7412&p=85791#p85791

That's where we put old deleted news and announcements that some people may still need to refer to.

---

Cpdn news

Mo.v and Les

Many thanks for your patience in your answers.
I have got it into my thick head now that there is no "Yardstick" for me to gauge my computer's progress except within the same app. My apologies for not realizing the variations between different apps. As I said, I am not concerned about how much credit is given, but just keep an eye on the slope of the line on the statistics page to indicate any problems that might be arising.

Thank again for the explanations.

Keith

Credits are useful for this purpose; a fall can indicate problems. If more members kept an occasional eye on things, posted when they notice a problem and persisted in asking questions until they understand how things work we wouldn't see CPDN pages like this.

---

Cpdn news

Woweeeeee!!! Incredible.

I cannot believe that 100% list of errors in computing.
Surely the guy must notice that something is wrong here.
He did get a few credits but nothing seemed to work for him after that.

If only he was not ANONYMOUS, a message could be sent to him/her.

Keith

There is a similar case at QuantumFIRE, also on a Linux box,but with a faster CPU. I believe those guys are trying to crunch too many WUs at the same time and get errors.I am using a very small cache (0.25 days) and get a new unit in 6 BOINC projects on my Linux box only when the preceding one has been uploaded at least. I rarely have an error. Only a hadcm3istd unit errored after 4600 hours run time and 4000 hours CPU time for reasons unknown. Overclocking can also cause errors. I have an Opteron 1200 CPU at 1.8 GHz and never overclock.
Tullio

---