Hello a@h-team,

I am receiving the stock- instead of the SSE2-application with my Linux Mint 15-PC (P4, BOINC 7.0.27). With Windows XP all works fine.
Should I use an "app_info.xml" or what can I do else?

Thank you for an answer

Dirk

With your 'Computers: hidden' I can not see how many 'plain' tasks did you have.
Are they more than 5?

It is normal for the server to 'probe' your computer with different apps to see which is best.
Look 'Number of tasks completed' and 'Average processing rate' at the bottom of my computer 'Application details' page (the 3 types of 'Period Search Application 102.10'):
http://asteroidsathome.net/boinc/host_app_versions.php?hostid=110

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

@BilBg

Thank you for the reply!
I opened my profile now and tried it again this morning after a reset. I actually got a SSE2-copy of period search with an estimated runtime of about 9.5h, which looks very better than the 24h yesterday.

Dirk

This question comes up quite often now.

When the project releases a new app version the server tries to learn which one works best for your machine. It does this by trying all the versions and comparing the run times. If you just let it do it's thing it will work it out without any intervention. There is no need to touch anything.

---

BOINC blog

well, it would be nice though if we could select manually, theres no point in wasting compute time while finding out which App works best if you know your CPUs capabilities.
I got the SSE2 App, though my CPU would do SSE3...

well, it would be nice though if we could select manually, theres no point in wasting compute time while finding out which App works best if you know your CPUs capabilities.
I got the SSE2 App, though my CPU would do SSE3...

There has to be a lowest common denominator, some people have no clue they just want to crunch. The units you crunch non optimally ARE still useful, juts like the ones you will crunch more effectively, it is NOT a waste of time and resources.

thats why there is an anonymous platform mechanism, just leave it as it is and give download-links for all apps

thats why there is an anonymous platform mechanism, just leave it as it is and give download-links for all apps

EEeewwww, manual, that's sooo 1930. Today we have computers to do complicated things like that for us. Just be patient and the computer will figure it out for you. If they give download links the links have to be maintained manually serverside as well as clientside. Experience has proven numerous times at numerous projects that inevitably either the admins or a volunteer will make a typo and provide/use a bad URL so that's far from being a fool proof approach. Also if users have to decide which app they want to run there will be many who pick the wrong one and run the slowest one rather than the the fastest one which would waste far more compute time than is wasted by the current method.

---

BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux

I got the SSE2 App, though my CPU would do SSE3...

In fact SSE2 and SSE3 apps are almost the same speed (very hard to judge if one of them is faster on my CPU)

... give download-links for all apps

http://asteroidsathome.net/boinc/download/

You can find yourself some posts with app_info.xml

But there is really no need.
See the number of tasks the server sent me for 'Period Search Application 102.10' (plain - only 3, SSE2 - 13, SSE3 - 951):
http://asteroidsathome.net/boinc/host_app_versions.php?hostid=110

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

very nice, thats exactly what I was looking for...so I guess I didnt look hard enough :)

EEeewwww, manual, that's sooo 1930. Today we have computers to do complicated things like that for us.

lol yeah, very complicated. That the devs of this project do not know a better approach than trial and error doesnt mean that its complicated.
The devs way it takes 40h of compute time to find out the correct app, doing it manualy takes about 10seconds.
Sure, people who do not know how to determine the app, the approach taken here is OK, though it would hurt no one to offer the other way too.

EEeewwww, manual, that's sooo 1930. Today we have computers to do complicated things like that for us.

lol yeah, very complicated. That the devs of this project do not know a better approach than trial and error doesnt mean that its complicated.
The devs way it takes 40h of compute time to find out the correct app, doing it manualy takes about 10seconds.

If you're suggesting the devs here are not too bright you're dead wrong. Kyong and HA-Soft are Czechs not goosesteppers and they're amongst the best you'll find anywhwere.

Sure, people who do not know how to determine the app, the approach taken here is OK, though it would hurt no one to offer the other way too.

OK, both approaches might be good but what if the app you would think should be faster is faster only for some tasks and the app you would think should be slower is the one that is faster for the "average" task. In that case the only way to determine the best app is to test or do what you mistakenly refer to as trial and error.

Also you are ignoring the maintenance issues that occur when using your approach. (Actually, I haven't RTFM'd on those issues for a while. I've never needed to use your approach and therefore have zero experience with it. I may be wrong but based on what I read in project forums there are maintenance issues. Correct me if I'm wrong but I'm a goosestepper too so... ;-))

On the positive side.. don't feel bad about missing the link to downloads because this project might not have one. The downloads directory and URL is a built-in feature of BOINC server and exists for every project that does a standard BOINC server installation. I suspect that's how BilBg knew the URL. Some projects choose to hide the link for whatever reason, for example they might not want volunteers to select the app on their own for the reason I mentioned above.

---

BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux

The devs way it takes 40h of compute time to find out the correct app, doing it manually takes about 10 seconds.

Don't be so sure.
Choosing the app only by higher CPU features (e.g. AVX over SSE3 over SSE2) is not always the best approach.
Testing the particular app build on the particular Hardware/OS/drivers/... combination is the real way to be sure.

AVX Asteroids@home app proved to Not be 'the best' for all CPUs/OSes that support it.

On SETI@home
- at present tasks and app builds: SSE2 is faster (by only 1-2%) than SSE3 app (tested on many CPUs/systems)
- in the past (different tasks and app builds): the SSE4.1 was slower than SSSE3 for most CPUs except Intel Core 2 Duo/Quad with very fast RAM
For i7 and i5 the best was SSE3 and Not SSSE3 or SSE4.1

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

I suspect that's how BilBg knew the URL

In fact it is written (for every project you use) in the client_state.xml

If you search for asteroidsathome.net you will find lines like:
<url>http://asteroidsathome.net/boinc/download/period_search_10210_windows_intelx86__sse3.exe</url>
<url>http://asteroidsathome.net/boinc/download/period_search_10210_windows_intelx86__sse2.exe</url>

<url>http://asteroidsathome.net/boinc/download/12f/input_21856_29</url>
<url>http://asteroidsathome.net/boinc/download/b4/input_21857_28</url>

... which promptly give you the hint for the general link.


And if you already have the files (period_search_10210_windows_intelx86__sse3.exe) in the project folder (asteroidsathome.net_boinc)
you don't even need to go and Download them, only save a copy (in case BOINC deletes them) and make an app_info.xml


P.S.
For other projects the links may be differently composed.

E.g. SETI@home have master URL:
<master_url>http://setiathome.berkeley.edu/</master_url>

... some other links like:
<scheduler_url>http://setiboinc.ssl.berkeley.edu/sah_cgi/cgi</scheduler_url>
<url>http://boinc2.ssl.berkeley.edu/sg_images/arecibo_181.png</url>
<url>http://setiboincdata.ssl.berkeley.edu/sah_cgi/file_upload_handler</url>


... and Download links like:
<url>http://boinc2.ssl.berkeley.edu/sah/download_fanout/225/31au13ab.24829.23337.438086664208.12.238.vlar</url>
<url>http://boinc2.ssl.berkeley.edu/sah/download_fanout/367/03au13aa.1155.476.438086664195.12.226</url>

(all info from client_state.xml)

... so the general link is:
http://boinc2.ssl.berkeley.edu/sah/download_fanout/

... but is not accessible - you have to know the name of the file to get it, like:
setiathome_7.00_windows_intelx86.exe
setigraphics_7.00_windows_intelx86.exe

... which gives links that work:
http://boinc2.ssl.berkeley.edu/sah/download_fanout/setiathome_7.00_windows_intelx86.exe
http://boinc2.ssl.berkeley.edu/sah/download_fanout/setigraphics_7.00_windows_intelx86.exe

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

P.S.

For SETI@home I see you use:
CPUID: Intel(R) Core(TM)2 Duo CPU E6750 @ 2.66GHz
SSSE3x Win32 Build 1846
http://setiathome.berkeley.edu/result.php?resultid=3410001998

Probably it (AKv8c_Bb_r1846_winx86_SSSE3x.exe) is better than AKv8c_Bb_r1846_winx86_SSE2x.exe

... but did you do a Benchmark:
http://lunatics.kwsn.net/index.php?module=Downloads;catd=5

You need for that 'MBbench 2.10' and 'V7 MB test WUs 1' (short WUs):
http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=387
http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=296

Also 'WisGen WUs.7z'
http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=384

From _WisGen_Notes.txt
"_WisGenA.wu adds planning for Autocorrelation so is meant for setiathome_v7."

My addition:
Make a copy of _WisGenA.wu and rename to _WisGenB.wu , put both files in TestWUs together with those from V7_MBtestWUs1.7z

in TestWUs you'll have:
_WisGenA.wu
_WisGenB.wu
PG1327_v7.wu
PG0009_v7.wu
PG0395_v7.wu
PG0444_v7.wu

... and you have to put apps (exe, dll) in:
MBbench\Science_apps\Reference\ (the SSE2)
MBbench\Science_apps\ (the SSSE3)

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)