464.25 RAC/host, how is this possible?

log in

Advanced search

Message boards : Number crunching : 464.25 RAC/host, how is this possible?

Author Message
Profile [SETI.Germany] Sutaru Tsureku (aka Dirk :-)
 
Send message
Joined: 28 Mar 10
Posts: 21
Credit: 101,159
RAC: 110
Total hours: 314,877
Message 837 - Posted: 6 Feb 2013, 8:47:47 UTC

http://wuprop.boinc-af.org/top_hosts.php

464.25 RAC, how is this possible?
79.88?
63.09?
59.78?
56.61?

I thought the highest possible RAC would be 56 / host.

24 hours / 3 hours = 8 WUs x 7 Cr. = 56 Cr. day = 56 RAC.

____________

Link
Avatar
Send message
Joined: 20 Jun 12
Posts: 36
Credit: 170,678
RAC: 0
Total hours: 79,093
Message 838 - Posted: 6 Feb 2013, 10:05:52 UTC - in response to Message 837.
Last modified: 6 Feb 2013, 10:07:24 UTC

Hmm... I wanted to post here something, but the system doesn't let me... says it's spam.

EDIT: Yeah... there was something weird going on, he was getting even over 20,000 credits per day.

EDIT2: now it worked to post exactly the same text when editing... stupid anti-spam system.
____________
.

Profile BilBg
Avatar
Send message
Joined: 20 Jun 12
Posts: 63
Credit: 80,489
RAC: 28
Total hours: 95,507
Message 839 - Posted: 6 Feb 2013, 18:40:06 UTC - in response to Message 838.
Last modified: 6 Feb 2013, 19:24:16 UTC


I think the anti-spam system works by comparing the length of links to the raw text
(you can't just post a link, you have to type enough text around it)

***

About this strange 'high-end' ;) system - I can only guess he found a way to make the clock tick fast?
Or compiled his own 'cracked' BOINC which gives to the apps fast ticks/time (180 times faster)
(will this lie the app that 1 minute is 3 hours?)

Other strange numbers for this ARMv7/Android system:
Tasks 0
... but:
Number of tasks today 1410

Even if this number 1410 is not updating (is not zeroed) unless host contacts the server
it means one task per minute for this host:
24 h * 60 m = 1440


____________



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

Link
Avatar
Send message
Joined: 20 Jun 12
Posts: 36
Credit: 170,678
RAC: 0
Total hours: 79,093
Message 840 - Posted: 6 Feb 2013, 22:16:26 UTC - in response to Message 839.

Other strange numbers for this ARMv7/Android system:
Tasks 0
... but:
Number of tasks today 1410

Even if this number 1410 is not updating (is not zeroed) unless host contacts the server
it means one task per minute for this host:
24 h * 60 m = 1440

No, the number is not updated as long as the host does not connect to server.

Maybe there should be an upper limit for tasks per day (10 or so), AFAIK BOINC supports that, Rosetta has for example a maximum of 100 tasks per day and core, you never can get more than that.
____________
.

Profile skgiven
       
Avatar
Send message
Joined: 7 Sep 10
Posts: 349
Credit: 806,146
RAC: 316
Total hours: 1,455,687
Message 843 - Posted: 7 Feb 2013, 17:49:59 UTC - in response to Message 840.

An anonymous platform is being used!

http://wuprop.boinc-af.org/host_app_versions.php?hostid=46285

Application details for host 46285

Data collect version 3 (anonymous platform, CPU)
Number of tasks completed 8482
Max tasks per day 9
Number of tasks today 1410
Consecutive valid tasks 0
Average processing rate 483.93999685775
Average turnaround time 0.00 days

For comparison (an x86 laptop):
Data collect version 3 3.40 windows_intelx86 (nci)
Number of tasks completed 33
Max tasks per day 43
Number of tasks today 1
Consecutive valid tasks 33
Average processing rate 0.093238493416704
Average turnaround time 0.39 days

There are a few funny systems out there. For example,
CPU type GenuineIntel Pentium(R) Dual-Core CPU E5200 @ 2.50GHz [Family 6 Model 23 Stepping 10]
Number of processors 3

It's not HT, http://ark.intel.com/products/37212/Intel-Pentium-Processor-E5200-2M-Cache-2_50-GHz-800-MHz-FSB

Profile BilBg
Avatar
Send message
Joined: 20 Jun 12
Posts: 63
Credit: 80,489
RAC: 28
Total hours: 95,507
Message 849 - Posted: 8 Feb 2013, 2:09:49 UTC - in response to Message 843.


So it's fake app that forges fake results (one result per minute).
If this 'app' goes to the wild (to be used by other credit hunters) its fake results will pollute the database.

Admin have to find a way to block such results server side (in validator?)
e.g. by comparing the times of 'Sent' and 'Time reported' (with some tolerance e.g. > 2.5 h)
(with 'Consecutive valid tasks 0' for this host maybe some similar change in the validator is already in effect?)

(the local (host) times are in the result file (e.g. wu_v3_1360245605_14634_0_0), maybe this fake app forges exactly them:

00:45:54 (6732): cycle1
...
00:46:54 (6732): cycle2
00:47:54 (6732): cycle3
00:48:54 (6732): cycle4
...
03:36:54 (6732): cycle172
03:37:54 (6732): cycle173
)


____________



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

Profile [SETI.Germany] Sutaru Tsureku (aka Dirk :-)
 
Send message
Joined: 28 Mar 10
Posts: 21
Credit: 101,159
RAC: 110
Total hours: 314,877
Message 857 - Posted: 8 Feb 2013, 20:01:13 UTC
Last modified: 8 Feb 2013, 20:16:14 UTC

The OS is Android, AFAIK for smart- & cell- phones, netbooks and tablets.
4 CPUs, to I guess netbook or tablet.

IMHO, 'anonymous platform' shouldn't be allowed here.
I guess it's possible to adjust the server software to allow only stock apps.

Someone informed the admin about this?

[EDIT of the title from 464.25 RAC, how is this possible? to 464.25 RAC/host, how is this possible?]
____________

Profile skgiven
       
Avatar
Send message
Joined: 7 Sep 10
Posts: 349
Credit: 806,146
RAC: 316
Total hours: 1,455,687
Message 863 - Posted: 9 Feb 2013, 17:36:26 UTC - in response to Message 857.
Last modified: 9 Feb 2013, 17:39:04 UTC

For several GPU projects it's distinctly advantageous to run more than one task at a time, using either app_info or app_config; 450K credits/day vs 880K at POEM, for example. Other projects include MW, WCG's HCC and on some rare occasions its also been more productive for GPUGrid. These setups, if correctly understood, should contain useful information for here. Note also that such setups usually mean a reduction in CPU usage for other apps, so there isn't normally an inherent advantage for here (it won't get you a higher colour of badge or more points).

This is a different situation because it's not a GPU project, the system is using "Anonymous platform (CPU)", and it's running on Android.

BTW there are '4-CPU' phones too, and an extra 400 Boinc points per day isn't a big deal; my RAC is ~800,000.

Link
Avatar
Send message
Joined: 20 Jun 12
Posts: 36
Credit: 170,678
RAC: 0
Total hours: 79,093
Message 864 - Posted: 9 Feb 2013, 17:42:59 UTC - in response to Message 863.

And what has all of that to do with this project, where we are allowed to run 1 WU per machine, which takes 3 hours and gives 7 credits?
____________
.

Profile [SETI.Germany] Sutaru Tsureku (aka Dirk :-)
 
Send message
Joined: 28 Mar 10
Posts: 21
Credit: 101,159
RAC: 110
Total hours: 314,877
Message 865 - Posted: 9 Feb 2013, 23:00:50 UTC - in response to Message 863.

skgiven, what happen at other BOINC projects it's up to the admins there and you.

WUProp collect data of other BOINC projects.

IMHO, WUProp should accept only the stock apps (of WUProp) - and no 'anonymous platform' (app_info.xml file usage), here at WUProp.

So noone can cheat here at WUProp.

hostid=46285:
Current Cr.: 49,350
and RAC: 464.25

All other WUProp members get max. 56 RAC/host.

I don't think that the above mentioned host collect useful data for WUProp.

____________

Link
Avatar
Send message
Joined: 20 Jun 12
Posts: 36
Credit: 170,678
RAC: 0
Total hours: 79,093
Message 866 - Posted: 10 Feb 2013, 8:48:53 UTC - in response to Message 865.

I don't think that the above mentioned host collect useful data for WUProp.

Even if it collected any data (I doubt that too, probably just fake data to get thru the validation), that data is not worth more than 56 credits per day, just like the data collected by any other host.

BTW, the other hosts which are just slightly over 56cr/day are probably OK, if you look at the tasks of this host, it has run some WUs without network connection (longer runtime, more credits), that can bump the RAC a bit. Besides of the RAC they all look OK to me and return the normal amout of tasks per day.
____________
.

Profile skgiven
       
Avatar
Send message
Joined: 7 Sep 10
Posts: 349
Credit: 806,146
RAC: 316
Total hours: 1,455,687
Message 868 - Posted: 10 Feb 2013, 12:41:52 UTC - in response to Message 866.
Last modified: 10 Feb 2013, 12:45:39 UTC

IMHO, 'anonymous platform' shouldn't be allowed here.


I wasn't sure what you meant by that; an anonymous platform used for WUProp, or you don't want any data taken from other projects that use anonymous platforms (possibly for the badges), or both.
Thanks for clearing up what you meant.

And what has all of that to do with this project, where we are allowed to run 1 WU per machine, which takes 3 hours and gives 7 credits?


Again, didn't know he meant 'anonymous platform' for WUProp. So I wanted to explain why I don't agree with excluding other app_info/app_config projects. After all, WUProp collects data about other projects no matter how they are run. WUProp might just see 'GPU app being run', rather than say 4 GPU tasks being run. So you don't get more hours, and lose CPU time. AFAIK WCG's HCC GPU app is recorded like this (they grant badges for hours of returned, validated tasks).

We are in agreement that the anonymous platform shouldn't be allowed with the WUProp app.

However this form of cheating is fairly limited (possibly to one individual). There are alternative ways to increase credit here too - using VM's, running multiple Boinc instances, and on hacked versions of Boinc.

The use of VM's is definitely going on here; people are running lots of VM's inside which they run Boinc with WUProp and other non CPU Intensive programs such as HAL. By doing this people can increase their WUProp credits, and HAL credits. It's fairly easy to run >10VM's per computer. Just look at the top users and how disproportionately high their credits are for here compared to other projects. My Boinc RAC is ~800,000 and my WUProp RAC is 333. Most of the top users also hide their systems. To have a WUProp RAC of >5000 you need 80+ systems. The 'top user' would need to have over 300systems running 24/7. What's the chances?

So while I agree with banning the running of the WUProp app in an anonymous platform, its a tiny (individual) problem and won't in itself level the playing field. VM's need to be considered too, and this is a much greyer area.

While I disagree with the use of many VM's (say 5+) running nothing other than WUProp (and perhaps HAL), I don't disagree with using one or two VM's; some projects are/have been Linux only (Asteroids), and some are mt (Simone) which interferes with GPU projects. The best/only way to run these if you have a Windows system or a GPU is in a VM. If you didn't run WUProp there would be little or no data for these projects. I expect some of the 'top users' collect little or no useful info for here.

Link
Avatar
Send message
Joined: 20 Jun 12
Posts: 36
Credit: 170,678
RAC: 0
Total hours: 79,093
Message 869 - Posted: 10 Feb 2013, 14:50:52 UTC - in response to Message 868.

We are in agreement that the anonymous platform shouldn't be allowed with the WUProp app.

Not sure... from what I have read in other threads, new apps or platforms are tested that way, IMO better limit max. tasks per day to maybe 9 or 10.

Also I'm not sure how helpful it is to have any information from VMs, the results of this project should tell people how each project performs on real hardware.

Is it not possible to restrict number of cores in use in case of mt applications? Or why are they interfering with GPU projects?
____________
.

Profile skgiven
       
Avatar
Send message
Joined: 7 Sep 10
Posts: 349
Credit: 806,146
RAC: 316
Total hours: 1,455,687
Message 870 - Posted: 10 Feb 2013, 16:09:36 UTC - in response to Message 869.

I agree that some info such as run times from VM's might not be very accurate; a native Linux client should slightly outperform a Linux VM. On the other hand, some data is better than none and the amount of RAM, HDD space required and file upload/download sizes should be the same.

I don't think it's presently possible to restrict/set the CPU core number to use for an mt app, even with recent beta Boinc clients which are trying to do just that.

In my experience, when I'm running GPU's apps and an mt app, when the mt app starts it uses all the CPU cores and stops the GPU app (which needs some CPU to run).

Profile Coleslaw
     
Avatar
Send message
Joined: 11 Apr 10
Posts: 157
Credit: 3,998,888
RAC: 1,713
Total hours: 6,785,548
Message 871 - Posted: 10 Feb 2013, 16:26:29 UTC
Last modified: 10 Feb 2013, 16:30:36 UTC

Well, maybe the developers need to come up with a way to distinguish if BOINC is running within a VM some how. It already detects if the OS has the software installed, but perhaps finding a way to identify if it is whithin one. That info is still helpful. I would rather run Linux from a VM then a multi-boot setup. I will not run it as a primary OS.

MT apps definitely interfere with GPU work as skgiven suggests. It can also cause some of your threads/cores to sit idle while waiting for them to open up if you run other projects. Some behavior I have noticed in Ubuntu was that if I downloaded the MT app while only 2 of 3 cores were open, it would only look for the 2. If it loaded with 3 cores available, it would wait and free up all 3. If I have long running work on a single core, it might be several hours before the other 2 cores do anything due to waiting on the third. BOINC wont start other single threaded apps in the mean time. Virtual machines help with this because I can just assign a set number of cores/threads to it while leaving the others to handle GPU tasks outside of the VM. WUProp can run in both to monitor my contribution.

zombie67 [MM]
         
Avatar
Send message
Joined: 30 Mar 10
Posts: 123
Credit: 4,929,457
RAC: 2,599
Total hours: 14,676,489
Message 876 - Posted: 12 Feb 2013, 0:57:44 UTC

WUprop tasks in VMs are completely valid. For example, some project are linux only. The only way I can run them is inside a VM. That other-project task is doing all the normal stuff it would otherwise, in a native environment. Using the same amount of RAM, same run-time, etc. The data returned to WUprop is just as valid, and worth the same credits, IMO.

However, none of that has anything to do with the problem that started this thread. The issue is that the project allows anonymous platform. While extremely helpful and necessary for other projects, I am not seeing the value here.
____________
Dublin, California
Team: SETI.USA

Profile skgiven
       
Avatar
Send message
Joined: 7 Sep 10
Posts: 349
Credit: 806,146
RAC: 316
Total hours: 1,455,687
Message 879 - Posted: 12 Feb 2013, 2:10:36 UTC - in response to Message 876.

I pulled the thread into the VM territory, for a reason; perspective. Running a hacked WUProp app in an anonymous platform is a problem, but relative to the many that are running dozens/hundreds of VM's with nothing other than WUProp and HAL attached, the single anonymous phone hack is almost irrelevant. Should it grow then so will the need to fix it, as was and is the unresolved issues of VM's!

BTW. I'm presently using one VM to run apps that are unstable in Windows, and one project that is just plain and simple faster. Another two uses to notch up for the VM.


Post to thread

Message boards : Number crunching : 464.25 RAC/host, how is this possible?


Home | My Account | Message Boards | Results


Copyright © 2017 Sebastien