I'm seeing some very interesting differences with the claimed credit that results from wu's processed on Linux vs. wu's processed under Windows.

First, the machines are near identical in terms of computing power. In fact, the cpu_run_time_pref reported by each is identical. The actual CPU seconds processing times and the cpu processing performances I see in my local logs are close enough to each other that they would be considered statistically equal. But for some reason the credit requested by the Windows clients is three (3) times larger then the Linux work units!

There are some differences between the Boinc clients (5.2.5 vs. 5.2.13) and the Ralph app version (4.84 vs. 4.92). But is this enough to throw the claimed credit differences that far off? My thought is that someone should look into this and come up with a reasonable explanation.

Here are some examples:
Windows:
WU 15378
WU 15380
WU 15381

Linux:
WU 15264
WU 16066

It appears that this issue may be somewhat more severe. I'm seeing a similar disparity with the Rosetta production version as well. I need to do some more analysis, but my quick look at results are showing a similar trend.

It boils down to 'boinc' and I wouldn't bother looking into it.

Given the times for the rosetta are near identical is good to know though.


Anyways,
These are the two important parts of the information to look at
Measured floating point speed
Measured integer speed
for the computer, it is that and purely that that time the time taken that gives you the points.

Windows is well known to give higher values due to the way it's compiled (boinc client) you can manipulate thoose values yourself if you want as you can compile your own client.
Have a read here may give you more insight.
http://boinc.truxoft.com/

also do a search on either of the forums for credit and there's been plenty of discusion on proposed ways to solve it.

---

Imho what causes windows claiming that too much higer than Linux
is that applications does *not* checkpoint enough.
Thus, after each suspend/resume a part of the WU is re-crunched again (hehe)

So, that excessive high claims are legit -:!
* and not caused by benchmarks in anyway.
-> after all, the work was done, and re-done and re-done ... and on and on

read here (I found this problem cause this occured with me)
http://boinc.bio.wzw.tum.de/boincsimap/forum/viewtopic.php?t=274

*I hope all apps checkpointing at least at each 10 seconds
to not waste cpu time re-crunching ... and to avoid that excessive high claims -:)

*such excessive higer claims, cause others to believe u are a cheater
and worse ... some projects as simap wasting about 2 teraflops
only to keep a quorum of 3 and grant the median_claim

---

Click signature for global team stats

I would have to look into it further to confirm, but thought I read somewhere that the way Linux counts CPU time vs. the way Windows counts it differs a bit. I'd have to check up on it again before I'd feel comfortable being quoted on what I'm only half remembering now...

If they're close in actual run times then this is a good sign, as the Linux variant has many times had lower performance... Einstein@home was facing that for a bit, with continual tweaks to the linux science app being made. Now that akosf has his optimized Albert app, and after some of his improvements get introduced into the official science app; I imagine this discrepency will increase on that project again. Well, until further work, allows some of his improvements to be ported over to the Linux variant of their science app also...

---

It is a known problem that the benchmark scores under Linux are about half of what they are under Windows. The run-time for the work units is the same though. This results in about half the credits being claimed.

--- bt