Seti@home gauntlet 2012

[lane42]

Thanks blckgrffn

 

[Pokey]

I'm down to a few GPU units. I'm out of CPU units.

According to the S@NL guys it is a Berkley server issue. Everybody is running out.

Yuck!

 

[wirelessenabled]

Speaking of Team Enterprise, Has anyone heard anything from Smoke lately??

Sorry I didn't find out about this Gauntlet until after the sign up was ended.

I am running my crunchers for Team Enterprise. I can't find a PM address for SmokeBall (Greg) anymore. I sent a PM to Crazee (Will) a few days ago and have not had a response thus far.

Also, after many years I am winding down on DC. My intention was to shut down after getting 10,000,000 on Seti, to go with our 16,000,000 on Rosetta, but I may let them continue until they each die.

Wirelessenabled (Jim)

 

[Sunny129]

I'm down to a few GPU units. I'm out of CPU units.

According to the S@NL guys it is a Berkley server issue. Everybody is running out.

Yuck!

yeah, i've been following it on the SETI@Home message boards. long story short, the consensus is that both uploads and downloads are severely backed up, which is consistent w/ what i've been experiencing in the last 18 hours or so...earlier my uploads were piling up while downloads were fine...and now, downloads are stacking up while uploads are fine...and of course this has to happen in the middle of the Gauntlet AND just before the weekend starts . though my cache is set to 10 days, it realistically only fills up to about 3 days worth of work for my GTX 460 and GTX 560 Ti. hopefully i'll make it through the weekend...

 

[biodoc]

My caches are full now. I check the transfer tab now and then to see if there are a backlog of downloads that have been labeled "retry" or "project backoff" and then select one and hit the retry button. This does work now and then.

 

[petrusbroder]

biodoc's procedure works often but quite a few times it does not.
If it does not, go in the BOINC-manager to the "Advanced" menu, click on "Do network communication now", return to the "Transfer"-tab and hit "retry now". That has worked very often for me ...

 

[biodoc]

Stats are out. We've solidified our #4 position but we've lost a bit of ground on taking #3.

 

[Sunny129]

well i was down to ~500 cached tasks this morning, but i'm almost back up to a full cache of 800 tasks now...and despite "average bits in" seems to be dropping rapidly in the SETI@Home cricket graphs, my uploads seem to be working fine too...

 

[lane42]

Stupid question time,
How are we in 4th place, but 3rd in output (Rac)

 

[Rattledagger]

Stupid question time,
How are we in 4th place, but 3rd in output (Rac)

RAC is Resent Average Credit, this is a weighted average that's exponentially decaying with half-life of 2 weeks.

So, if example you've been getting 100 k/day for 1+ months time, your RAC will be 100 k/day. If you completely stop, after 2 weeks your production would be 0 k/day, but your RAC will still be 50 k/day.

During this gauntlet it's the other way around, long-time active SETI-users has roughly the same RAC and production, while someone that once-upon-a-time ran SETI but haven't for a long time started with a RAC of roughly zero and it takes some time before RAC has increased.

Now, it's a good chance makes an error here, but for someone starting the gauntlet with RAC roughly zero and procduces 100 k/day, they'll maybe hit RAC 100 k/day after 2 weeks, meaning at the same time the gauntlet finishes.

 

[petrusbroder]

... and that is why RAC is not such a great way to compare production in the short run.

 

[Pokey]

yep,
I'm up to 60,000 + production, but my RAC is still in the 30,000 range.
So We are going faster than we look...........

But I don't think it will be enough to get us out of 4th place. I for one should have started earlier to take into account the ramp up time, and learning curve and also, we could have used a bunch more GPUs............ :\

At least I did get Lunatics installed, which helped.

I'm still enjoying the race......... Just wish it could be for another week or two.

 

[Pokey]

Gauntlet has just been updated.

Hmmm......... the scenery isn't changing............

For the team or for me.

 

[Sunny129]

...I for one should have started earlier to take into account the ramp up time...

i'm in the same boat...had i had all of that insight, i would have started ramping up 2 weeks ahead of time...

 

[blckgrffn]

Gah, GT 240 rig has 295 drivers, couldn't figure out why CUDA WU's kept pausing.

Monitor goes to sleep, CUDA does too. Lame. Didn't have access to it over the weekend, so it's fixed now.

http://setiathome.berkeley.edu/forum_thread.php?id=67043#1198322

WHQL isn't like it was, I miss the Good ol' days... like when you had to install different drivers for different games No, I don't miss those days at all but it peeves me when little things like this get by either AMD or nVidia QA.

I'll be going to for 1M regardless of when the race ends

About getting ramped up - that's what's nice about the PrimeGrid races. They have the infrastructure in place to handle really shallow queues and they only count work downloaded and uploaded during the race time. I have finally figured out to get my rigs going a day or two early, but even a couple hours before/after the race starts is OK... with this, we probably need to be on it a week ahead of time or so, minimum?

 

[Sunny129]

About getting ramped up - that's what's nice about the PrimeGrid races. They have the infrastructure in place to handle really shallow queues and they only count work downloaded and uploaded during the race time. I have finally figured out to get my rigs going a day or two early, but even a couple hours before/after the race starts is OK... with this, we probably need to be on it a week ahead of time or so, minimum?

it would seem so...its one thing when a race is based only on work downloaded and uploaded during the span of the race, like PrimeGrid races...but if SETI races (or at the very least, this particular Gauntlet) are based on RAC as Rattledagger pointed out, then someone who hasn't participated in S@H for quite some time (and therefore has a S@H RAC of zero), will require 2 weeks minimum of ramp-up time just to make his/her RAC equal to his/her actual max production.

 

[blckgrffn]

Ha, well, I fired up SETI for the first time for this race and didn't want to convert everything over (ie, divert it from my other projects) until the race started.

A mistake, to be sure. A misunderstanding on how this worked vs my other limited participation in DC races at a minimum

 

[Sunny129]

you're not the only one...while i realized i would have to start ramping up before the actual race, i only started 2 days before the start of the race, not knowing that it would be based on RAC...had i known that, i would have started a full 2 weeks ahead of time since my project RAC was, at the time, zero.

 

[Rattledagger]

you're not the only one...while i realized i would have to start ramping up before the actual race, i only started 2 days before the start of the race, not knowing that it would be based on RAC...had i known that, i would have started a full 2 weeks ahead of time since my project RAC was, at the time, zero.

It's NOT based on RAC, that SETI@NL also includes RAC as part of their stats doesn't mean they're using this to decide who's getting whathever place in the race. They're normally using RAC as a measure for how many days to overtake, or be overtaken by, someone else, but they've mentioned multiple times on their international forum that the RAC is incorrect during this Gauntlet so isn't a good indication of anything.

As for ramping-up ahead of time, since SETI@home uses a quorum of 2 this can have some effect, since part of the work crunched before the gauntlet can be validated during the gauntlet, while part of the work crunched during the gauntlet won't be validated until after the gauntlet. While SETI has upto 2 months deadline depending on wu, there's also a limit of 50 wu/core or 400 per GPU, so despite these long deadlines most wu will still be validated within a couple days, and ramping-up a long time ahead of the start will therefore likely have little effect on the gauntlet-stats.

Then it comes to SETI@home, connection-problems to the servers and servers going-down has likely a bigger effect on the gauntlet-results than how many days before the start users started to run SETI@home.

Oh, and even in the unlikely instance SETI@home only delivers VLAR-wu's for the rest of the race, meaning especially the GPU user-group will take a performance-hit since Nvidia-cards won't get these wu's, it's still doubtful if any of the other teams would manage to catch them now, since the other teams also have multiple Nvidia-crunchers...

 

[biodoc]

It's NOT based on RAC, that SETI@NL also includes RAC as part of their stats doesn't mean they're using this to decide who's getting whathever place in the race. They're normally using RAC as a measure for how many days to overtake, or be overtaken by, someone else, but they've mentioned multiple times on their international forum that the RAC is incorrect during this Gauntlet so isn't a good indication of anything.

As for ramping-up ahead of time, since SETI@home uses a quorum of 2 this can have some effect, since part of the work crunched before the gauntlet can be validated during the gauntlet, while part of the work crunched during the gauntlet won't be validated until after the gauntlet. While SETI has upto 2 months deadline depending on wu, there's also a limit of 50 wu/core or 400 per GPU, so despite these long deadlines most wu will still be validated within a couple days, and ramping-up a long time ahead of the start will therefore likely have little effect on the gauntlet-stats.

Then it comes to SETI@home, connection-problems to the servers and servers going-down has likely a bigger effect on the gauntlet-results than how many days before the start users started to run SETI@home.

Great points rattledagger! :thumbsup:

Yes, forget about RAC. It's all about the points you generate during the race. Since there is a quorum of 2 for each WU, I try to jump in 3-4 days before the race with my hardware strategy in place (100% GPUs for the most part). Oddly enough my GTX295 box has out performed my dual GTX460 box and my GTX 570 box with my dual 9800GTX+ box trailing the others. Good thing I didn't sell the GTX295.

Also, I try to ramp up my cache to 2-3 days early on to prepare for the seti server down times. I believe in the last race in 2009, there were server issues on both of the weekends.

My total points for the race (7 days to date) is 601,000 and my RAC is 44,000. RAC is useless as a measure of productivity for those of us that jump into the race with a RAC of 0.

 

[lane42]

With your computers biodoc, you would be well over 100k :thumbsup:

 

[Pokey]

Looks like there is some bumping and grinding going on further back in the gauntlet but we seem to be stuck in 4th..............

We need to go out and pick up 8 to 10 GPUs apiece and maybe we could change the scenery...........

 

[dajeepster]

i'm as dry as a two dollar whore at a gang bang in the desert.... no wu here... and i'm set up for 10 days..

 

[biodoc]

With your computers biodoc, you would be well over 100k :thumbsup:

My 2 fastest computers (Sandy bridge) are running linux so if I had a steady diet of AP AVX WUs, yes I would expect that production level. When outdoor temps dropped to normal levels in the New England last Friday, I did engage the cpus but AP WUs are far and few between.

 

[Sunny129]

even though the server has been up and running for hours now since its weekly maintenance outage and is producing new work, the scheduling server is currently down. i monitor the Server Status page and the Cricket Graphs to try and make sense out of why uploads or downloads might have stopped...


*EDIT* - the scheduling server has been back up for some time now, according to the server status page. and while it seems like new work isn't flowing, i have confirmed that new work is slowly seeping out the splitter servers (i'm getting 1 new task at a time, even though i need ~200 more to fill the cache). also, it is good to keep in mind that the information on the server status page should be taken with a grain of salt. for instance, according to the server status page the upload server goes up and down several times a day at what seems like very short intervals (on the order or 15-30 minutes), however i never seem to have a problem uploading - even when the server status page says that the upload server is down. at any rate, i think the cricket graphs are a better indication of whether or not work is flowing to and from the server, and how much.