RNA World

[yoyo]

I checked this 189,000s workunit http://www.rnaworld.de/rnaworld/work...p?wuid=1015952 from LAURENU2 list. You see on the workunit page that this one is estimated with 318,000s runtime on the reference system. There is already one result finished for this wu, which needed 206,000s.

I checked also this http://www.rnaworld.de/rnaworld/work...p?wuid=1015891, which you aborted after 111,000s. This is estimated with 402,000s (4d 15h) on the reference system and finished by one host after 117,000s.

Yes, they are long, but not 10 days.

Maybe your Boinc DCF (duration correction factor) is to high and therefore the remaining time is so high. This time anyway is mysterious calculated by Boinc client.

yoyo

[LAURENU2]

Yes Yoyo I did abort 1 or 2 that were 90% complete and a run time of 50 & 95 hrs
It has to do with the promise I made to myself read above

The 10 days came from the est time of a WU listed here that I aborted
!0 days or 1 day the same looping thing can happen without Save points

What is holding you back from having SAVE POINTS
It Seems like most all other BOINC have them
You just tell it to save every 10 Min

I know nothing about a (duration correction factor) Boinc is a stock install with no tweaking

[Michael H.W. Weber]

I checked this 189,000s workunit http://www.rnaworld.de/rnaworld/work...p?wuid=1015952 from LAURENU2 list. You see on the workunit page that this one is estimated with 318,000s runtime on the reference system. There is already one result finished for this wu, which needed 206,000s.

I checked also this http://www.rnaworld.de/rnaworld/work...p?wuid=1015891, which you aborted after 111,000s. This is estimated with 402,000s (4d 15h) on the reference system and finished by one host after 117,000s.

Yes, they are long, but not 10 days.

Well, I see. His machines really picked the biggest ones that remained in the system (and will remain in the future as well for analyses of other organisms).

What is holding you back from having SAVE POINTS
It Seems like most all other BOINC have them
You just tell it to save every 10 Min

No, you do not 'just tell it to save'. If you want checkpoints (i.e. 'save points') then you need to write (or re-write) the entire code of your science application de novo. I have explained this many times (even in our FAQ). Most scientific applications that run on high performance clusters are not desiged to have checkpointing. With RNA World, we run such software and it is technically impossible to change the code to write the checkpoints: it would have to be re-written completely with a different design. Moreover, it is a waste of time to do this for all the implemented and the many, many upcoming software modules in the future. I know other DC projects have checkpoints, but these usually have one single science core client and the entire system was from the very beginning designed to run as a DC system - that is a completely different starting situation compared to what we are dealing with at RNA World. Hence, we thought about methods to develop a universal checkpointing, i.e. 'writing save points' at the system level to apply it to exactly these applications which do not allow for checkpoing generally - and these applications are undoubtedly the majority. This poses a different problem, namely that you require admin rights to save the entire RAM content to disk (it would be as if you send your laptop to sleep mode). Would you grant admin rights to a DC project? No. See, and that's the problem. We now have decided to use a virtual machine approach within BOINC. But that project is not yet complete.

I know nothing about a (duration correction factor) Boinc is a stock install with no tweaking

Yes, and as this stock install it has problems known for long. One is this DCF issue which means that the run time estimate BOINC displays in the form of your progress bar is adjusted by (your stock install) BOINC manager depending on how much your machines spends on tasks different from the BOINC tasks. If you play a game while RNA World computes, this gaming is at cost of RNA World WU progress. BOINC detects that and re-calculates the remaining run time. As such a neat idea. In practice, however, when you stop playing, BOINC takes a long time to re-adjust this correction (factor). As a result, it displays run-time estimates that are far too high. And this is exactly what happened with you RNA World WU. One could say that BOINC actually 'betrayed' you in telling you the wrong run time estimate.

Well, I think it should all be a bit clearer now. For me it is clear that running RNA World on machines that do not run 24/7 is a problem as long as we do not have checkpointing. And please note that I am by no means 'angry' or so because you have put forward some criticism here. The only thing I can do about this is trying to explain why we do the things we actually do and why some things are not yet implemented.

Michael.

[LAURENU2]

Well thank you Michael
That was a vary clear and explanation of how RNA works within BOINC
And I know you are not upset Nor am I upset It is just that I can't Do Long WU's
And was worried about the 8/5 user's stuck in a Loop

As for the DCF only 3 PC's out of the 60 I run here are used by people the 58 are only doing DC 24/7
And the 3 do not run RNA because they need a reboot from time to time

I will continue to Run RNA but I will keep a eye out for long WU's and send them back to you Like this one

2473870 1059802 12 May 2010 21:06:46 UTC 17 May 2010 13:36:43 UTC Aborted by user 327,886.09 296,718.66 1,285.62

I know thats a lot of points to flush but my word is my honor

[Michael H.W. Weber]

http://www.rnaworld.de/rnaworld/forum_thread.php?id=51

Michael.

[Michael H.W. Weber]

Hi guys, so the conference turns out to be a remarkable success. We have met a lot of people that are very interested in our project. Even better, we have acquired new collaboration partners that can help us in experimentally validating our computetaional results on a broader scale. Also, posters presented by independently working groups indicate that we are absolutely on the right track with our analyses. So, I am quite confident that we will soon make it for the first publication in a scientific journal. Moreover, we have scheduled the next conference presentation of RNA World which will take place in September in Dresden/Germany. Details on that at a later time.

Michael.

[Michael H.W. Weber]

We have released an OSX client today.

Michael.