2019 Formula Boinc Sprints

[Ken g6]

I blame all the flower badges at RakeSearch. They must all be putting out a No Tasks Scent.

 

[StefanR5R]

Any ideas which are best for what?
AVX--------intel?
SSE3------Ryzen?

Intel Haswell and later should run AVX2.

AMD Zen are meant to run the AVX2 application too, because Daniel compiled the current one specifically without the PEXT instruction which previously diminished performance on Zen.

 

[StefanR5R]

@Orange Kid, you will possibly need to be patient when you pull stats on the final day of the race. Last year, database performance of the RakeSearch server was severely lacking, such that the Formula Boinc server collected the final stats much later than we were used to from other sprints.

 

[StefanR5R]

I guess North America is now increasingly switching computers towards RakeSearch: Since an hour or so, I am experiencing quite many HTTP errors.

 

[crashtech]

For me it is an exercise in frustration. I need to further refine my methods to a more hands-off approach.

 

[Ken g6]

I finally figured out the limit for tasks in progress is 16 per core. So if you want to get more you're likely to be frustrated.

 

[Orange Kid]

I'd be happy with one per core at all times.

 

[Ken g6]

It figures, the race time rolls around, I get a few WUs uploaded, and then the site goes down before I can refill my cache.

 

[crashtech]

Yeah, this project is so not suitable for a Sprint, why does this keep happening? It sours me on the whole FB concept.

 

[Ken g6]

I gotta find a better way to babysit these BOINC clients. Scheduler requests for new work won't happen while either a download is not active or an upload is active.

 

[StefanR5R]

It figures, the race time rolls around, I get a few WUs uploaded, and then the site goes down before I can refill my cache.

Yeah, this project is so not suitable for a Sprint, why does this keep happening? It sours me on the whole FB concept.

The RakeSearch server becoming unavailable as soon as the sprint starts was, obviously, among the likely scenarios to happen --- given what we experienced at the RakeSearch sprint last year, and seeing how a few of the stronger teams appear to be even more active this year.

I gotta find a better way to babysit these BOINC clients. Scheduler requests for new work won't happen while either a download is not active or an upload is active.

(an upload -> too many uploads)
Yep, the need to retry stalled downloads before triggering an update is something which I have yet to work into my own toolset as well. Unfortunately it seems boinccmd is not particularly easy to use for this. This is perhaps a point at which scripting should be abandoned in favor of source code modification... But I for one am not equipped with the necessary spare time currently.

 

[StefanR5R]

Wow, formula-boinc.org was finally able to pull first non-zero race stats.
Current order: XtremeSystems > [H]ard|OCP > SETI.USA > SETI.Germany > TeAm AnandTech > Dutch Power Cows.

SETI.Germany aren't actually active in Formula Boinc, but many of them run RakeSearch as their team's project-of-the-year.

Edit,
taken from server_status.php just before the sprint:

Research progress

Rake search of diagonal Latin squares of rank 9 (%) ________ 75.581

I'm curious what the percentage will be when the sprint is over.

Edit 2,
Probably not much further, because the server evidently can handle only a fraction of what the volunteers could complete.

 

[Ken g6]

Hey, some stats came in! We're in 5th right behind SG.

 

[StefanR5R]

My E5-2696s went idle two hours ago, and my E5-2690s went idle now too.

I will leave the computers on their own over night, which is during the next seven hours. After that, before I have to leave for work, I will go and look whether the DoS situation has resolved itself. If not, I will switch my computers off.

After work, I will have a brief opportunity to check once more. Then I will be away for travel.

Since I am currently not able to micro-manage my computers like I used to, and since German electricity rates are far too high to let the computers idle for extended periods, I can't do better than this; sorry.

--------
(client not requesting new tasks if a download is stalled or if too many uploads are in progress)

This is perhaps a point at which scripting should be abandoned in favor of source code modification...

This cannot be worked around with scripts:
- Right now, stalled downloads stall again immediately when a retry is forced.
- A script can't effectively do anything against the ever growing backlog of pending uploads.
So, source code modification would be the only remedy against the client's refusal to request work. But what good would this do when the downloads don't succeed then.

 

[crashtech]

This project is a joke.

 

[biodoc]

I'm almost out of work.

 

[DooKey]

This project is a joke.

 

[Ken g6]

This project is a joke.

Well, at least I got my badger:

I imagine the badger badge was given to me by the badger badge badger. Too bad I didn't get either of the mushrooms.

 

[StefanR5R]

During the entire night, neither uploads nor downloads succeeded. But quite exactly at the time when I woke up, the first few transfers began to move again, and now there are even a few new tasks being processed.

at least I got my badger:

I got the Ladoga Seal of Approval. But I couldn't marvel over it yet, because the image file transfer failed.

 

[Kiska]

But I couldn't marvel over it yet, because the image file transfer failed.

This shows just how unable everyone is able to access the server

The gaps in the graph is munin timing out. And the timeout is set pretty generously, 60 seconds to wait for a response
Also located here: http://munin.kiska.pw/munin/Munin-Node/Munin-Node/index.html

 

[Ken g6]

The admins did something to the site and I think it's almost normal now. I still don't think it would handle the Pentathlon, though.

Edit: https://rake.boincfast.ru/rakesearch/forum_thread.php?id=146 if you can access the site.

 

[StefanR5R]

The RakeSearch server (or rather, the server which hosts RakeSearch in a VM) is a rather old one:
https://www.seti-germany.de/forum/threads/8907-RakeSearch?p=309780&viewfull=1#post309780
Looks like Nehalem or Westmere to me. I bet it has got an ancient mass storage system too.

During the entire night, neither uploads nor downloads succeeded. But quite exactly at the time when I woke up, the first few transfers began to move again, and now there are even a few new tasks being processed.

This worked for mere 3 hours, then things went downhill again and my hosts remained almost entirely idle.

The admins did something to the site and I think it's almost normal now.

Did they? Then maybe boinc-client's exponential backoff policy prevented my hosts to recover on their own. I am forcing some retries now.

.
.
.

Still lots of transfer failures. Hrmph. I need to leave now, but I leave the computers running. (I wish I had a script ready to run now which cuts off too large transfer backoff periods. Something to turn to later; should turn out useful in many more Formula Boinc sprints to come.)

This project is a joke.

To be fair, it is just one of so many projects which simply are not suitable for Formula Boinc sprints, given the current high participation in these sprints.

 

[crashtech]

@StefanR5R , right of course, I do tend to lead with my emotions at times. I was looking forward to putting my "new to me" 2P rig to good use, and we got RakeSearch instead...

 

[biodoc]

All my rigs are busy again.

 

[StefanR5R]

Mine were at 700 W when I came home from work, and are still below 1100 W now (should be above 1300 W). They are large machines with currently only a single client per machine, hence they are slower to recover.