Posts by xii5ku

1) Message boards : News : Power Outage Saturday October 7th (Message 4223)
Posted 22 Dec 2018 by xii5ku

Brandon,
thank you very much for taking care of this. Much appreciated.

2) Message boards : News : Power Outage Saturday October 7th (Message 4199)
Posted 1 Nov 2018 by xii5ku

@Brandon,
I have been tracking our user stats and team stats using boincstats:
https://boincstats.com/en/stats/63/team/detail/141/lastDays
https://boincstats.com/en/stats/63/user/list/0/0/141/0
(Following from the second page, there are also last-40-day records of each user.)

The result is here:
https://i.imgur.com/QXW97HJ.png
https://pastebin.com/raw/YqKhjK4n
In the PNG, the blue section shows team stats straight from your database, the black section shows user stats from your database and then added up, and the red section shows the difference between black and blue.

Notes:
- Since boincstats does not display fractional credits, there are small rounding errors in this spreadsheet.
- From what I understand, the differences between user stats and team stats during the most recent few days are merely caused by timing differences between user stats dumps and team stats dump, and are averaged out across days.

Evidently, the team stats are now being updated correctly again. Thank you very much for fixing this!

But also, 5 million points are missing in our team stats from the ~2.5 weeks with the database glitch. This is a lot. And it means a lot to us. (Speaking for myself, I don't care much for my user points at MindModeling, but I do care a lot for my team points.) Are you able to correct the team credits retroactively? (I have seen other projects' admins perform similar corrections when they had mistakes in their databases.)

Thank you.

PS,
as far as I know, TeAm AnandTech has been the most productive of your contributing teams in 2018 so far. And if we continue our effort (and could get our stats corrected), we shall make it into your all-time top ten very soon.

3) Message boards : Number crunching : New tasks with very short deadline (Message 4186)
Posted 17 Oct 2018 by xii5ku

Update: I am receiving tasks with 5 days deadline now. So, this looks like described in these posts from January 2016:
https://mindmodeling.org/forum_thread.php?id=1024&postid=3638
https://mindmodeling.org/forum_thread.php?id=1037&postid=3639

Brandon wrote:

Hey guys,

Here is how the work unit deadlines work currently.

By default our work units will now have a 5-day turn around time.

However, the work units at either the very beginning, or at the end of a batch of work, will have a < 1-day turn around time.

< 1-day turn around at beginning of the batch of work:
This allows us to get the modeler important information regarding their model faster, just in case a problem arises, and they need to stop the job early.

< 1-day turn around at end of batch of work:
This allows the job that is almost done to finish quickly, and get initial results on new work earlier.

The bulk of the work units in the middle will stay at the 5-day turn around time, and hopefully this fixes the issues you guys are having.

This current job is at the beginning so that's why the work unit deadline times are short but it will not stay this way throughout the entire job.

Let me see what I can do to help fix this and I will get back to you guys.

Thanks for supporting us and happy crunching,

Brandon

4) Message boards : News : Power Outage Saturday October 7th (Message 4185)
Posted 17 Oct 2018 by xii5ku

Badges and credits for users and teams are indeed strong motivators to many volunteers. This is what they were invented for, I guess.

@Brandon, thank you for keeping us informed. Get well.

5) Message boards : Number crunching : New tasks with very short deadline (Message 4184)
Posted 17 Oct 2018 by xii5ku

Hi,
there is a batch of "ACT-R cognitive modeling environment using Clozure Common Lisp (Linux Only) v2.35 (sse2)" tasks being issued right now which has a deadline of only 7 hours, not the usual 5 days. Is this deliberate or by mistake?

(edit) I am just noticing from a 2015/ 2016 thread that past batches occasionally had similar short deadlines, but not intentionally.

6) Message boards : News : Power Outage Saturday October 7th (Message 4166)
Posted 10 Oct 2018 by xii5ku

The stats for TeAm AnandTech are not tallied up correctly, it appears.

Team points update for 2018-10-10: +0 points, and 2018-10-09: +5,780 points. This is shown by both free-dc and boincstats. Their current total points of TeAm Anandtech, 3,980,049, are the same as those seen on the team page at mm.org right now.

Yet if I add up the team's individual user points at free-dc, I get for 2018-10-10: +52,360 points, and 2018-10-09: +173,330 points (+225,690 in total).

Adding the team's individual user points at boincstats, I get for 2018-10-10: +247,696 points, and 2018-10-09: +20,357 points (+268,053 in total).

Is this a normal glitch which will correct itself over time, or is something not working right anymore since the outage?

(I spot-checked another team, and this had less than 1 ‰ difference between team points and the sum of user points.)

7) Message boards : Number crunching : Native Java 1.7 Application requires 1.97 KB/sec download bandwidth... (Message 4068)
Posted 3 Nov 2017 by xii5ku

We have done a couple of deployments over the past week or so. During 2 of the deployments, the system was unavailable for less than ten minutes. You may have been trying to upload results at that time which is what reduced your BOINC Client's bandwidth value which is causing this issue.

This is *not* the cause.

The problem with bogus bandwidth measurements is a general one, caused by your project server. Probably linked to (a) your network connection being chronically swamped, (b) by far most of the transferred files being merely a few bytes small.

The problem occurred on *all* clients which I ever used at MindModeling. On a 22C/44T client it occured several times a day. On a 6C/12T client a few times a week. On a 4C/8T laptop roughly once a week. I had these on different internet connections, and fairly good ones.

The problem is solely on your end, and it is a *permanent* one. (Well, the repeated posts about it here should already tell you so.)

If you reset the project, that should reset which will fix this for you.

It is not a fix, only a temporary workaround until it happens again. But it is so far the only known working workaround.

If not we will look into other possibilities as to why this happened.

Can you disable bandwidth limits? You don't need them anyway; the only time when networking bandwidth is used is when a client starts working on a new dataset/ new application. After that, you distribute almost only those few-bytes transfers, whose transfer performance depends on networking latency (and maybe on *your* networking bandwidth), but not on the client's networking bandwidth.

Or if you cannot disable this limit, then you must not measure networking bandwidth of the small transfers, you should only measure large transfers.

Edit:
I.e. I suppose the solution is either in the boinc server config, or in boinc server program code (which would be troublesome).