(archive)MISFIT

[swl551]

The utility is done: MigrateGIOMSENT2csv


Here is what you see on startup of the utility

GIOM_SENT file aggregator v 1.0
Run this from main MISFIT directory where GIOM_SENT is a subdirectory.
Output file will be UPLOADED_RESULTS.CSV
Overwritten everytime this utility is executed.
PRESS ANY KEY TO CONTINUE

get from http://www.mersenneforum.org/misfit/

Future enhancement to MISFIT will build on this.

[JMLX]

Originally Posted by kracker

Hmm, before it escapes my mind, is there to balance *only* specific workers?
For example only, I have 2 projects I run, first proj. on workers 1 and 2, the second on workers 3 and 4.
Is there a way so I can balance without some of them "leaking"
Or simply, balance only specific workers as... said above?

Thanks!

Just run two instances of MISFIT. You can rename the exe so it allows two instances to run.

Set up each to manage the grouping you want to "relate"

[kracker]

Originally Posted by JMLX

Just run two instances of MISFIT. You can rename the exe so it allows two instances to run.

Set up each to manage the grouping you want to "relate"

I believe(?) MISFIT doesn't allow more than one instance. Hmm. maybe misfit1.exe?

[swl551]

Originally Posted by kracker

Hmm, before it escapes my mind, is there to balance *only* specific workers?
For example only, I have 2 projects I run, first proj. on workers 1 and 2, the second on workers 3 and 4.
Is there a way so I can balance without some of them "leaking"
Or simply, balance only specific workers as... said above?

Thanks!

Kracker has defined the scenario where workers/instances would be grouped together and the balancing of work would occur inside that group. Does anyone else have interest in this feature or additional ideas on how it could work?

[flashjh]

Originally Posted by swl551

Kracker has defined the scenario where workers/instances would be grouped together and the balancing of work would occur inside that group. Does anyone else have interest in this feature or additional ideas on how it could work?

I've thought about it before, and I could use it sometimes. I thought maybe you could put checkmarks/checkbox next to each worker in the list and only the checked workers would get balanced. That would work well for a once-in-a-while scenario.



Edit: Just finished analysis of the data looking for duplicates. The "duplicate" from the 17th, well, I can't find it. Maybe it was a glitch? Here are the results from all the duplicates I found:

FF	EXPONENT  FR	TO	UPLOADED
0	59508727	72	73	10/24/2012 6:53
0	59508727	72	73	10/24/2012 17:44
0	59733463	71	73	10/29/2012 16:06
0	59733463	71	73	10/30/2012 8:28
0	59794937	72	73	10/31/2012 19:20
0	59794937	72	73	11/1/2012 20:38
0	59881543	71	73	11/4/2012 22:34
0	59942557	71	73	11/5/2012 11:19
0	59881543	71	73	11/5/2012 13:26
0	59943703	71	73	11/5/2012 20:35
0	59945363	71	73	11/6/2012 3:54
0	59942557	71	73	11/6/2012 5:58
0	59947549	72	73	11/6/2012 9:06
0	59949437	72	73	11/6/2012 14:20
0	59952029	72	73	11/6/2012 19:15
0	59953687	72	73	11/7/2012 0:27
0	59955251	71	73	11/7/2012 8:48
0	59943703	71	73	11/7/2012 9:51
0	59956709	71	73	11/7/2012 17:14
0	59958421	71	73	11/8/2012 0:42
0	59959829	71	73	11/8/2012 9:07
0	59930539	71	73	11/8/2012 9:07
0	59962057	72	73	11/8/2012 14:19
0	59945363	71	73	11/8/2012 14:19
0	59964173	71	73	11/8/2012 22:45
0	59930539	71	73	11/8/2012 23:48
0	59965519	71	73	11/9/2012 7:05
0	59947549	72	73	11/9/2012 9:11
0	59966947	72	73	11/9/2012 12:22
0	59939107	71	73	11/9/2012 14:30
0	59968343	71	73	11/9/2012 20:46
0	59969753	71	73	11/10/2012 4:04
0	59939107	71	73	11/10/2012 4:04
0	59949437	72	73	11/10/2012 4:04
0	59937413	71	73	11/10/2012 6:11
0	59971621	71	73	11/10/2012 12:29
0	59973031	71	73	11/10/2012 20:49
0	59937413	71	73	11/10/2012 23:04
0	59952029	72	73	11/10/2012 23:04
0	59974489	71	73	11/11/2012 4:25
0	59975681	71	73	11/11/2012 12:54
0	59953687	72	73	11/11/2012 18:11
0	59977237	71	73	11/11/2012 20:20
0	59951587	71	73	11/12/2012 0:35
0	59979133	71	73	11/12/2012 4:46
0	59980721	71	73	11/12/2012 12:04
0	59951587	71	73	11/12/2012 15:13
0	59983277	71	73	11/12/2012 20:31
0	59985383	71	73	11/13/2012 7:20
0	59955251	71	73	11/13/2012 7:20
0	59992157	71	73	11/13/2012 19:50
0	59993377	71	73	11/14/2012 4:33
0	59994661	72	73	11/14/2012 9:43
0	59990641	71	73	11/19/2012 3:37
0	59981377	71	73	11/19/2012 3:37
0	59990641	71	73	11/19/2012 4:39
0	59981377	71	73	11/19/2012 4:39
0	59969753	71	73	11/20/2012 16:30
0	59974489	71	73	11/20/2012 16:30
0	59971621	71	73	11/20/2012 16:30
0	59966947	72	73	11/20/2012 16:30
0	59977237	71	73	11/20/2012 16:30
0	59965519	71	73	11/20/2012 16:30
0	59959829	71	73	11/20/2012 16:30
0	59964173	71	73	11/20/2012 16:30
0	59980721	71	73	11/20/2012 16:30
0	59975681	71	73	11/20/2012 16:30
0	59985383	71	73	11/20/2012 16:30
0	59994661	72	73	11/20/2012 16:30
0	59968343	71	73	11/20/2012 16:30
0	59956709	71	73	11/21/2012 11:53
0	59958421	71	73	11/21/2012 11:53
0	59993377	71	73	11/21/2012 11:53
0	59979133	71	73	11/21/2012 11:53
0	59983277	71	73	11/21/2012 11:53
0	59962057	72	73	11/21/2012 11:53
0	59973031	71	73	11/21/2012 11:53
0	59992157	71	73	11/21/2012 11:53

From the analysis, I can see that the problem was getting worse for some reason (and I was wasting a lot of time!). After the new version came out and I started uploading once per day, the problem immediately disappeared and has not happened since. Also of note, none of the duplicates were exponents with factors. I wish I knew why it appeared that there was a duplicate the other day? I don't have any .html log files for the old duplicates since the version on the Nov 21st started those. The logs from 17th - 18th don't show the duplicate either, so who knows?

As for the .csv tool, very helpful. Since it was the first time I put it to use, I also compiled the .txt files myself and did a separate analysis on the duplicates - both versions were a perfect match.

Edit2: The FLASH report is perfect!

[swl551]

First, guess who I named the "FLASH" report after...

Anyway I don't think we'll ever get the exact answer here. The most likely culprit was over-active work balancing, but even under controlled testing we couldn't make it produce duplicates.

I think the key here is that the program has been improved to provide prevention, detection, forensics and analysis tools.

Since the first pubic release to the latest, hundreds of hours of work has been spent improving things.

If there are any future oddities hopefully we'll have everything in place to get accurate answers.

[swl551]

See screenshot on how balancing work between "selected" instances will be implemented.

Considering the minimal interest for "instance grouping", and the complexity of building an "instance group" manager, I think the ability to "select and balance" sort of meets in the middle and provides the desired outcome.

[flashjh]

Originally Posted by swl551

See screenshot on how balancing work between "selected" instances will be implemented.

Considering the minimal interest for "instance grouping", and the complexity of building an "instance group" manager, I think the ability to "select and balance" sort of meets in the middle and provides the desired outcome.

I think that will be a nice feature.

[kracker]

Originally Posted by swl551

See screenshot on how balancing work between "selected" instances will be implemented.

Considering the minimal interest for "instance grouping", and the complexity of building an "instance group" manager, I think the ability to "select and balance" sort of meets in the middle and provides the desired outcome.

Yay! That will work very nicely for me atleast

[kladner]

That looks great!

[swl551]

VERSION 2.2.6
1. New feature allowing Direct Selection of instances to balance work between. Just click the instances in the grid you want to balance and then right click. Use CTRL key to select non contigious blocks, and SHIFT for contigious blocks. (standard windows keys). This feature is independent of auto-balancing or clicking the "Balance All" button. Depending on your needs you may use all, none, or just one of the three.

2. Minor tweaks.


Get from http://www.mersenneforum.org/misfit/