Boinc pentathalon race

[Fardringle]

It looks like most of the TAS folks are participating in this race as the monthly "project".

It's difficult/inconvenient for me to get my computers to change projects reliably once a month. Even using Boincstats BAM, I occasionally have to manually stop and restart the BOINC clients in order to get them to pick up on new projects. There's no way I could get them to switch every 5 days and I definitely can't control when they send in results. So I think I'll just stick with YoYo until I hit 1 million credits later this week, then work on a few more milestones for the rest of the month, or maybe switch to whatever is the "race" project at that time and work on it until the end of the month...

 

[Ken g6]

Weird!! FreeHal uses next to nothing on CPU cycles. It looks like you can crunch it at the same time as a CPU intensive project.

Weirder:

FreeHAL is in fact a CPU intensive application that is just configured to sleep most of the time to make it "virtually" non-CPU intensive...because we think that we can reach more users, when it can run beside other projects without disturbing by using much ressources. For users who want Freehal to use more CPU, there is the possibility to change the number of work units running at the same time.

FD: N/P This definitely isn't a race for everyone. If they don't pick PrimeGrid as the math project, I may just quit halfway through. :\

 

[Ken g6]

Well, that's interesting. Our XML file is "not well-formed"!

I sent an email to get this problem resolved.

Note: I just checked with wget, and it looks like it's my Firefox that's having parsing problems, not their server. :$

 

[Rattledagger]

I sent an email to get this problem resolved.

Doing a little testing, the problem is the & in the url...

Seeing it's using an older url to these forums, the easiest would be to just email the founder, and get him to update the url to the new url after the forum-changes...

Hmm, the founder is listed as TAC, so pm to the relevant parties should be enough.

 

[filibusterman]

This method won't really work with Milkyway, since Milkyway limits you to 48 tasks at once per GPU. On the 58xx-series of clients, this means less than 2 hours cached work.

As for Einstein@home, it seems the "Arecibo"-files finally has started to download, so while I won't start running Einstein@home before evening the 4th., I should atleast have the work cached beforehand.

are we restricted to arecibo binary or can we do the other apps associated with einstein...I hope the latter or I have a lot of not so helpful WU to contribute!

 

[Rattledagger]

are we restricted to arecibo binary or can we do the other apps associated with einstein...I hope the latter or I have a lot of not so helpful WU to contribute!

There's no mention of any limitations in this race for which of the Einstein@home-applications you can run, so run whatever you're getting.

But, while you can run any application, atleast according to the status-page, it can look like they'll run-out of S5GCE-work during the race. They're working on releasing the next "run", S5GC1, but it's not certain if this will be released immediately, or if there will be a gap. So, there is a possibility there'll only be Arecibo-work for part of the race, if you've not already grabbed enough work already.

Anyway, if the Einstein@home-wu's doesn't run much faster than estimated, I've got enough cached for the race, and I've now disabled network-connection on all computers. So, as they're finishing their current task, they'll start running Einstein@home instead.

Now, the next to remember in this race is to remember to enable Milkway on Saturday...

 

[biodoc]

For those of us on daylight savings time, this race starts at 8PM EST tonight, correct?

Ok, I admit it, I'm UTC challenged!

 

[filibusterman]

For those of us on daylight savings time, this race starts at 8PM EST tonight, correct?

Ok, I admit it, I'm UTC challenged!

I believe thats is what I also concluded from : http://setiathome.berkeley.edu/utc.php

 

[Rattledagger]

For those of us on daylight savings time, this race starts at 8PM EST tonight, correct?

Ok, I admit it, I'm UTC challenged!

This really depends on which "EST" you're talking about...

... since EST = UTC + 10 hours
and EST = UTC - 5 hours

In any case, the race starts in 4 hours 40 minutes.

 

[Ken g6]

And shouldn't it be 8PM EDT? :sneaky:

 

[Philippart]

hmm, I was buffering a huge number of WUs and crunched them till I noticed that recent boinc versions automatically report when the software notices that you're connected to the internet... lol

so no big WU queues ahead for me...

 

[filibusterman]

hmm, I was buffering a huge number of WUs and crunched them till I noticed that recent boinc versions automatically report when the software notices that you're connected to the internet... lol

so no big WU queues ahead for me...

 

[Rattledagger]

hmm, I was buffering a huge number of WUs and crunched them till I noticed that recent boinc versions automatically report when the software notices that you're connected to the internet... lol

so no big WU queues ahead for me...

Well, v6.10.xx basically works like v5.10.14 as far as reporting work goes, so it's "recent" as in 34 months old.

Now, there's many various reasons why work is reported. The 4 most common ones, and most likely to show-up during this race is:
1: If client needs to ask for more work, it will also report any finished work. This happens if estimated cached work drops below your cache-settings (the sum of "connected..." and "additional...").
2: If a task has been "ready to report" for 24 hours.
3: If time to deadline is less than your "connected..."-setting.
4: If user manually hits "update".

For step 1-3, disabling network by setting "Network activity suspended" will make sure nothing is transferred, until user manually enables network again.

For #4, if user manually hits "update", BOINC-client expects the user knows what he's doing, so will open-up network-connection for 5 minutes, even if it's currently disabled.


BTW, the trickly part can be later in the race, depending on which projects is choosen in the two remaining disiplines, it can be possible you'll want to start running one of these, and at the same time run Milkyway. Due to Milkyway not allowing large cache of work, you can't really disable network while you're running Milkyway...

edit - Just took a look on the progress on my Einstein-work, it seems to progress faster than estimated, especially Arecibo-work is running in half the estimated time, so even with 6 days cached I'll need to cache more work...

 

[Ken g6]

You don't need to cache enough work to get to the end of the race; just the start which is in ~2.5 hours. You might want to move on to another project even before time runs out on Einstein.

 

[biodoc]

And shouldn't it be 8PM EDT? :sneaky:

I stand corrected!

I'm still not sure when the race starts!:biggrin:

Since I'm crunching "offline", I'll wait until I see the stats start rolling in. :whiste:

 

[Rattledagger]

You don't need to cache enough work to get to the end of the race; just the start which is in ~2.5 hours. You might want to move on to another project even before time runs out on Einstein.

You don't need to cache beforehand, but with Einstein@home using a quorum of 2, it is a bigger chance the work will be validated during the race if you do grab all work now, instead of re-filling on the last day.

Since new projects is announced 7 days before their start, and each project will run for 5 days, chances are there's either a project starting 13. and one the 14., or both remaining projects starts on the 14th.

So, if math-project isn't Collatz, there will be 2 cpu-projects overlapping on atleast 4 of the 5 days... This can lead to an interesting finish, depending on which project each team chooses to put majority of their resources on...

If math-project is Collatz, there's maybe a 1-day GPU-overlap with Milkyway, but bio/med will in this case run as only cpu-project at the end of the race.

In any case, if haven't mis-read any of the strange rules, the 2 remaining projects should both be announced within 48 hours from now...

 

[Ken g6]

You don't need to cache beforehand, but with Einstein@home using a quorum of 2, it is a bigger chance the work will be validated during the race if you do grab all work now, instead of re-filling on the last day.

You're right! I didn't think of that! :thumbsup:

In any case, if haven't mis-read any of the strange rules, the 2 remaining projects should both be announced within 48 hours from now...

Right again. And the math project is Collatz. And I have no compute-capable video cards. :\

Although I am curious how GPUs compare to CPUs in Collatz. Is one high-end nVIDIA (GTX 275-285) roughly equivalent to one quad-core CPU? How does ATI compare?

 

[Rattledagger]

Right again. And the math project is Collatz. And I have no compute-capable video cards. :\

Although I am curious how GPUs compare to CPUs in Collatz. Is one high-end nVIDIA (GTX 275-285) roughly equivalent to one quad-core CPU? How does ATI compare?

No idea how much faster GPU's are compared to cpu, and I've never had a nvidia so even less sure on this.

They announced Collatz later in the day than expected, but seeing on the other projects these was also announced at midnight UTC, so should have expected it. With Collatz starting a day earlier, there's 2 days overlap with Milkyway, so the time teams switches can play a role...

In any case, based on the end-date, the last project should start the 14th, and this will be a cpu-project.

 

[biodoc]

We are currently in 12 place. stats

Here's a bit more info on Milkyway and Collatz:

CPU WUs: I've never crunched these projects on my CPUs. Can someone else provide info here?

GPU WUs:

Collatz: quorum of 2, minimum requirement is single precision compute capable ATI or NVIDIA card. ATI cards outperform NVIDIA.

Milkyway: quorum of 1, minimum requirement is double precision compute capable ATI or NVIDIA card. ATI cards outperform NVIDIA by a larger margin.
Here's a list of cards that work with Milkyway.

Since Collatz requires a quorum of 2, I would get your single precision cards crunching a few days before start of collatz section of the race since those cards will not work with Milkyway.

Since all 3 of my cards are double precision, I will keep them on Milkyway for the full 5 days and then switch to Collatz.


FreeHal: On the project preferences page, you can assign up to 25 WUs to crunch simultaneously on each rig. From my brief testing period, you can not build a cache of WUs and the quorum for WUs is 1. EDIT: I can't wait to see how much CPU is used crunching 25 WUs simultaneously (5% ?)

 

[rabrittain]

...haven't bathed or shaved. I've been in my underwear all day. I'm jacked up on caffeine and a monster energy drink, and I'm wondering why a couple of my einstein work units won't upload.

 

[rabrittain]

OK ... I've come back down a little.

I have elected to seek the advice of those who understand this race better than I do. All my Einstein work units have reported; I have downloaded as many Milky Way work units as I'm allowed. I've turned off network activity. Am I correct?

 

[biodoc]

We are now in 14th place.

Here's some individual stats for day one:

 

[dajeepster]

I've got one computer down with a bad pump.... I'll pull the two 8800gts out of it this weekend and popped into another computer til I can get the pump fixed/replaced.

 

[biodoc]

Nice crunchin' everyone!! We moved up to 13th place!

The stats will look better tomorrow.

Dajeepster, your AC must be set on high for this race!

 

[zzuupp]

You're right! I didn't think of that! :thumbsup:



Right again. And the math project is Collatz. And I have no compute-capable video cards. :\

Although I am curious how GPUs compare to CPUs in Collatz. Is one high-end nVIDIA (GTX 275-285) roughly equivalent to one quad-core CPU? How does ATI compare?

On an ATI 4850, Collatz tasks have been finishing in 1,000 - 1,100 seconds with a few outliers. In round numbers, let's say that's 17 minutes. There are no CPU tasks left in my history. IIRC, those took many hours.

I'm not the only one who has been running this & hasn't checked in on this thread. Hopefully, the rest will continue to running Collatz as well.