for last 7 completed Wus.
claimed granted (newest completed WU at the top)
61.21 54.58
65.08 54.27
64.58 43.91
62.83 40.43
65.17 67.09
64.03 68.73
65.24 66.77
Can someone advise why 'granted' has fallen so much recently? PC runs more or less 24/7. Run time was changed but before any of the above WUs.

---

CPU type AuthenticAMD
AMD Athlon(tm) 64 Processor 3200+
Number of CPUs 1
Operating System Microsoft Windows XP
Home Edition, Service Pack 2, (05.01.2600.00)
Memory 959.48 MB
Cache 976.56 KB
Measured floating point speed 1855.84 million ops/sec
Measured integer speed 3401.8 million ops/sec

-----------

Some of the things I'd ask are whether it's just AMD Windows users that are noticing a decline. I'd look at my results, but my Rosetta system seems to be offline at the moment. :(

When the granted credit system was first changed, I was averaging a nice bonus; although a few months later, that bonus had decreased. If lots of users are noticing a recent decrease in the claimed/granted credit ratio - it could be because of the mix of cpus and OSes being used by the recent influx of new users. (Lots of Linux users would decrease the average.. and a major change in the intel/amd cpu could also be responsible - if this is a project wide issue.)

On the other hand, if it's just your machine, a change in anti virus software (Norton's bloatware) can cause a loss of cpu cycles. (I noticed a slight improvement switching to Panda Titanium). Virus/Trojan/Worm infections can also cause a loss of cpu cycles and system slow down.

Run through Spybot Search and Destroy, Adaware, and spywareblaster. Next, use TrojanHunter 4.6, Ewido which is now called AVG Anti Spyware 7.5. Then do a couple of online scans (www.bitdender.com, www.pandasoftware.com, www.trendmicro.com, etc.) That'll catch most of what's out there and help rule out malware.

---

i'd expect the increased number of contributing CPUs that don't give much higher benchmarks but have a lot higher rosetta throughput will result in the reduced granted credits. e.g. the larger caches on newer CPUs won't affect the benchmarks, but will increase rosie throughput - resulting in lower average claims for a given WU.

I think there's another factor involved too though - the change has been quite quick, but as you say, an influx of linux/mac machines could do this...

---

System archetecture probably has more to do with this than a problem with the credit system. The credits you listed are from different work unit types. Some of those types might be an especially big calculation, something that doesn't fit in cache on AMD but does on Intel (just guessing, it could be related to other platform differences).

If that's happening, the Intel cpu's would be doing more 'work' in the same period of time since the AMD systems are having to hit system memory. This would skew the credit calculation downard for AMD processors since it would take AMD longer to do each simulation.

---

Bennyrop - have been running avg7.5 for several weeks now. No Norton stuff on this PC!
Housing - hope that 950mb approx mem is sufficient for rosie, also have 2.something gb virtual mem.

The change between claimed/granted credits was so sudden. Previously claimed/granted was approx equal over time.
There have been no system changes that could create such a disparity.PC is not yet 6 mos old.

---

Michael, pull up your chart (statistics tab) in your BOINC client for the host average. You haven't actually said if your credit per hour of crunching has been effected, or it you just notice the credit granted vs. claimed differing.

Also note, about 4000 new ESL users joined in the passed two weeks. ESL is an online gaming organization... so probably lots of high-end machines at work there. So if they yield similar benchmarks, but larger numbers of complete models, then machines not completing as many models will get less credit.

---

Rosetta Moderator: Mod.Sense

[quote]Housing - hope that 950mb approx mem is sufficient for rosie, also have 2.something gb virtual mem. [quote]

Cache and system memory are a bit different. Cache is much smaller and 'closer' to the CPU. AMD cpu's have less cache than intel, so, if it is related to cache then amd cpus are going to spend more time waiting for information from system memory.

My theory is that the new work units that came out on 4/2 are cache hungry and the intel cpu's are pushing amd scores down. . . and I just ordered 87 dual amd workstations today :)

The gaming folks could also be impacting things, although they've been around a couple weeks now.

---

I got between 70 and 80 per workunit consistenly with an RAC of 225+. I the past two weeks, these have also seen the same reduction of granted credit and RAC fell to 205. I did a complete re-install to lean the system and the down time reduced the RAC to what you see now. Downtime was three hours. (easy to loose; hard to get back)

System is visible for all to see. No hardware changes and task manager shows 99+% CPU on rosetta over a five minute period.

The Rosetta team is checking for any code inefficiencies that may be causing the lower scores.

System runs 24/7.

I though that scores were based on what like-systems were getting on RALPH and not on what a dissimilar gamming system was scoring?

I thought the new credit system was based on a running average of all the WUs returned for that particular type of WU (job?). So if the big gaming systems are running down the average everyone gets less.

Let's start the credit granting discussion again. That was fun :)

---

Proudly Banned from Predictator@Home and now Cosmology@home as well. Added SETI to the list today. Temporary ban only - so need to work harder :)



"You can't fix stupid" (Ron White)

Weird - my latest completed WU
claimed credit = 62.69
granted credit = 64.96

must have been doing something wrong!! OR - have all the ESL pples gone home?

---

Our claimed/granted numbers are bouncing all over as well. There seems to be very little stability or logic behind the numbers. It's too bad Rosetta can't develope a credit system to equal the high standards it achieves with its science and level of Admin communication. I'm the last person to suggest that the credit can of worms be opened again but IMHO Rosetta is losing crunchers because of it....:(....

At a huge risk (and I'm kinda paranoid bringing this up), the pervious credit discussion last year involved the BOINC benchmark credit system (benchmark & CPU Time) Where if a machine benched higher, it would score higher.

This seems to be the opposite. If your machine completes much more work than the benchmarks would indicate then you score higher. (I get it now).

If the gamming machines are getting more work done then that's great!

However, if your machine does the same ammount of work as it did three weeks ago, Then your scores go down instead of remaining the same.

I just hate to see anyone's contribution be less valued for the same amount of work as before.

Fingers crossed as I click "Post".

on the WU's listeed below I took a beating on points.
there were no errors and everything ran full time of 8 hrs +/-
why would this happen? my machine was idle for most of this time.

WU# 63345082 C)70.88 G)54.12
WU# 63240173 C)77.96 G)46.84

---

on the WU's listeed below I took a beating on points.
there were no errors and everything ran full time of 8 hrs +/-
why would this happen? my machine was idle for most of this time.

WU# 63345082 C)70.88 G)54.12
WU# 63240173 C)77.96 G)46.84

A hammering nails analogy. . nails are each simulation (decoy). You 'pay' your workers by how many nails they hammer. Intel has a slightly bigger hammer than amd, if you're working with small nails both processors can hammer them in with a single stroke. Let's say you get a little bit bigger nail for a new job, just to the point it takes amd two strokes to hammer it in while intel can still do it with one. That little change had a noticeable effect since now the amd guy is getting 'paid' less.

The little change in this case is probably the type of work unit. It seems like they're studying some fairly complex stuff, so I would expect this to happen from time to time.

---

...great analogy

...and if a CORE2 duo benchmark predicts that a new RNA nail can be set with 3 strikes, and it actually only takes two... and another system also predicts it will take 3 strikes, and this proves to be true... In the end your credit claim gets thrown in to the average with everyone else's, and your credit granted is based upon that rolling average.

So, if a higher mix of CORE2 processors (just one example, it might be large L2 cache, it might be fast memory bus, and if it's not one thing, it may be next task) comes on the scene, you get lower credit claims thrown in to the average then you would have otherwise. Because their claim is based on the 3 strike prediction provided by the benchmark. But the CORE2 gets higher credit then that, because everyone else reporting in is taking longer to do the same work.

What I'm attempting to emphasize here (not very well I'm afraid) is that the potential for reduced credit granted has nothing to do with the raw speed of a CORE2. It has to do with the benchmark of a CORE2 as compared to the actual work it can perform, and how that compared to another machine's benchmark as compared to it's actual work performed.

As you've probably noticed, even though the BOINC benchmark remains the same, the figures it produces vary over time. And Rosetta models on a given task do not all require exactly the same number of CPU cycles to run. Some models start looking good and so Rosetta spends more time investigating them. Others start to look poor and Rosetta decides that particular model is not worth the time to investigate further. This variation is expected, it's part of the science, and it averages out across time. You are just as likely to get a long (or short) running model as anyone else. And there is no means of predicting ahead of time which models will be long (in fact, if that could be done, then the non-interesting models would have been eliminated prior to sending the tasks out).

---

Rosetta Moderator: Mod.Sense

Housing and Food Services is correct in that some of the current experiments have more variable run times per decoy and are also taking longer to run. For the jumping runs, we haven't added more frequent checkpoints as with the ab relax runs but we are working on it. Also, there are cases during the jumping runs where chain breaks have to be closed and this part is sometimes inefficient so we'll have to optimize this step. hang in there and we'll work out the kinks that inevitably arise with the high pace of new experiments.

I would like to thank the Rosetta gurus for their usual timely and incisive answers to the credit wobbles.....however, the bottom line is still the same: if it's not possible to fairly accomodate AMD crunchers (as an example) with this credit system, then Rosetta will lose them to other projects......

Is there no neutral way of determining credit or make a average point scale of amd and intel processors so that no one gets shorted on credit?

I would like to thank the Rosetta gurus for their usual timely and incisive answers to the credit wobbles.....however, the bottom line is still the same: if it's not possible to fairly accomodate AMD crunchers (as an example) with this credit system, then Rosetta will lose them to other projects......

---

Is there no neutral way of determining credit or make a average point scale of amd and intel processors so that no one gets shorted on credit?

The credit system is neutral, amd and intel are getting the same number of credits per decoy returned (by work unit type). It just turns out that Intel has the faster equipment (or larger cache, prefetcher, or whatever that causes them to be better on this given wu type) and is able to return more decoys per unit time. As the scientists write new code testing new theories, they might try a calculation that turns out to be better on AMD hardware.

---