DPC cause Dnet proxy blockage! (Anyone surprised?)

[AppleTalking]

Here's decibel's latest .plan:

"The master is back-logged right now, and has been for quite some time.
(http://n0cgi.distributed.net/rc5-proxyinfo.html) This is due to a
'MegaFlush' that the Dutch Power Cows are doing, despite our requests that
they discontinue this practice.

The effect of this backlog is that daily stats will be off for the next
several days. Everyone's stats will appear low for today, and will slowly
return to normal over the next few days.

Outbound work will still be available. Once the backlog has been processed
by the master, RC5 stats will return to normal. OGR stats should be
un-affected. Expect the occurance of dupes to increase while this backlog
exists as well."

The master has been getting repeatedly blocked up for the past several weeks, but now it's up to 11 million and counting! Geez, I hate megaflushes. Give me the slow and steady keyrates . . .

Note to DPC members: I really don't have anything against the DPC, I'm just posting this as an FYI. I love friendly competition!
.

Nick Klingaman
appletalking@home.com

 

[Poof]

:Q

What is it? A full moon or something?

SETI is hard down, the dnet stats are now going to be pretty much down, my folding@home stats haven't been updated in like a week...

Sigh...

(/me does have genome@home stats updating though... )

Thanks for the info AppleTalking!

 

[Russ]

What does this mean?



<< Expect the occurance of dupes to increase while this backlog exists as well. >>



Lost production because of this?

Russ, NCNE

 

[Dijkhuis]

Moose:

<< Oh one other thing... Those of you who save up for flushes should flush ASAP. once we reject work, if you have work with v2.8012 even if it was finished before we start rejecting, you will not get credit for it in stats. >>



Maybe that wasn't really smart after all.

 

[ZapZilla]

Someone post a link to this on the DPC forums, so the DPC can
justify their irresponsible actions.

I've always been against the DPC's teamwide organized mega dumps.

The DPC should be ashamed of themselves.

What is DNet's plans for upgrading the master keyserver?

What computer is the master keyserver currently?

 

[ViRGE]

HUmm, all I can really say towards the DPC's is shame on you. Dnet's told you before, more than once I belive, not to dump. Even if Dijkhuis is right, and you only dumped becuase of the client problems, I find it dissapointing to say the least, for your dump today. You shouldn't have had to dump for any good reason(getting to 1bil may have been the exception); so for your sake with Dnet , and your reputation; I must offically ask that you forever end these mega-flushes. They look cool, but as long as Dnet's not prepared for them, you're doing harm to everyone else too.

PS I also hope for your sake that this was just an overactive mini-team, and not DPC wide

 

[Kilowatt]

Wave a finger and scold them, when they sould be getting the boot.

I hope they are proud of themselves again.

That's my opinion, and I'm sticking to it, so flame on

 

[Napalm381]

The DPC discussion thread on the matter is here. Apparently, Moose said they could flush it. Oh well....

 

[networkman]

Simple solution: Flush the cache(read DELETE!) and re-issue the keys. Problem solved!

That will eliminate mega-flushes very quickly, regardless of who is doing it!

Flame on, folks. That's my opinion, and YES, I am involved in the RC5 project too.

 

[crYnOid]

I don't think that moose would have known the amount of work that the dpc had stored up. If they had less stored this wouldn't be a problem.

Anyway it has happened, and it's time for me to get to uni.

 

[Possessed Freak]

The sad truth is that as long as DPC members continue to hold blocks for individual dumps, when everyone is asked to dump (as moose said) the team will do a 'mega flush'. TA does it to, when a miniteam or individual wants to be high in ranks for the day, they save up. Combine that with 10 or so of these savings and then asked to dump, wham thats a lot of blocks.

Short answer to the problem: get better servers because people will always save up for a higher stat for a day.
Longer answer: perhaps run a DISTRIBUTED server solution. set up win2k adv server or something where only dnet reps have admin rights to the box and lock down all resources to it. That way TA could run their own keyserver, process their own blocks, and have less computing power needed at dnet themselves. I bet somebody out there would fund this because 'dell' or 'compaq' servers would be running the 'worlds largest computer'.

 

[dennilfloss]

Apply to megaflushes the same penalty as negative marking on a multiple-choice exam. If you megaflush, a - sign will show in front of your daily total and your blocks for that day will be subtracted from your total.

 

[Viztech]

See that thread that Napalm posted.

Moose TOLD THEM to go ahead and flush. I guess that the DPCs are still guilty of hoarding though..

viz

 

[Choralone]

I checked my e-mail after not being able to get on the board (high traffic errors) and saw the notice from decibel. What gets me is why don't the DPC just dump everything on a daily basis? Why the hoarding of blocks constantly? They've been warned time and time again not to do it. It's one thing to dump 250K blocks in one day for a PTE like we've done in the past but anything over a million more than your &quot;normal&quot; daily TEAM keyrate is just unacceptable. Just dump it all everyday, or every other day, and be done with it!

 

[Moose]

Well the whole conversation was not posted.

I was asked what I would do since the 8012 clients were getting marked as bad. In that conversation I told them I would flush in order to not lose work. However I also stated that we have asked them before not to store work and create mega flushes. Consistantly they have ignored this request and stored up for a mega flush. reguardless if I had told them to flush or if they had flushed 2 weeks later, the same effect would have happened.

My major concern when talking to them was to inform them that they might loose work if they continue to hold work back to create a mega flush in the future. The way I look at it better to warn them of the lose of blocks due to continueing to hold work than to not tell them anything at all.

As I said before the entire context was not stated. I was asked what I would do and then if our network could stand up to the workload placed on it by another mega flush due to the need to flush work saved up. Well truth is yes it can. The network performed as expected. If more work comes in it is held until it can be read by the master server. The result however is something that is not desired by others who do not hold work for the purposes of mega flushes. That result is a backlog. Work does not show up in stats on the day it is flushes. Sure normally you might flush work at 2300UTC and it doens't show up until the next day stats. But now with a large backup we have seen work not show up for days. This causes more users than normal to say things like &quot;I'm missing 100,000 blocks!&quot;

The flush has occured. The backlog is already going down. I would suspect things will get back to normal as far as the master is concerned by morning. Lets just put this one behind as moose's fault and get on with life. Lets also take this as an example of why we have asked in the past that mega flushes not occur.

Thanks
paul

 

[eNeRGy]

<< I've always been against the DPC's teamwide organized mega dumps. >>



This wasn't a organized DPC teamwide mega dump.



<< PS I also hope for your sake that this was just an overactive mini-team, and not DPC wide >>



It were a couple of teams, not that many. And for my sake, WE DON'T ORGANIZE THEM ANYMORE SINCE SEPTEMBER. *sigh*




<< The sad truth is that as long as DPC members continue to hold blocks for individual dumps, when everyone is asked to dump (as moose said) the team will do a 'mega flush'. TA does it to, when a miniteam or individual wants to be high in ranks for the day, they save up. Combine that with 10 or so of these savings and then asked to dump, wham thats a lot of blocks. >>



Finaly someone who understands it. This is not DPC doing this, a lot of theams are saving up blocks. DPC does sometimes lose their daily #1 spot, and AT is losing sometimes their #2 spot due to smaller teams which are saving up.



<< The flush has occured. The backlog is already going down. I would suspect things will get back to normal as far as the master is concerned by morning. Lets just put this one behind as moose's fault and get on with life. Lets also take this as an example of why we have asked in the past that mega flushes not occur. >>



We we're not responsible for this alone. This is coincidence of things. And I really don't like us as team to be blamed in a .plan. This was an action of members individually, not a team action

 

[Russ]

Looking at the stats for the 27th, the impact appears to be minimal. Maybe the problem will be larger in the 28th's totals, but we (TA Cube) appear to be down only a few K from where we should be for today.

Russ, NCNE

 

[JollyP]

Blame the DPC, thats easy. The statsoutage from early december 2000 which lasted more than a week was our fault too offcourse.

Why not blame D.net for not keeping up hardwarewise with the growth of the their own project or releasing clients that are not tested well enough?

As far as organised MF's are concerned, there hasn't been one since last september. Only individual members and miniteams are saving in order to get in our daily DPC hitparade or to break a DPC miniteam flushrecord but thats their choice and has nothing to do with planned DPC MF's.

My 2 eurocents, have a nice day all

 

[Onno]

<< Consistantly they have ignored this request and stored up for a mega flush. reguardless if I had told them to flush or if they had flushed 2 weeks later, the same effect would have happened. >>


Absolutely not.

It are individual 'subteams' of DPC that are saving up blocks, these blocks were never meant to be flushed all at once.

It really bites me that decibel publicly is blaming DPC for the lack of stats, only because a few of our members are flushing some saved up work. DPC members are not the only ones saving, nor the only ones flushing, we just are the largest team doing that.

But what do you expect? We are the largest team, by far... More than 5% of the world-wide daily RC5 production comes from DPC. Of course we will also be the largest contributor of blocks when a thing like this happens.

(and I do have my questions about the whole backlog thing... in January something like 8M extra blocks were flushed, on a total of more than 40M that's only 20% more, did that really have to cause a 40M backlog?)

 

[mechBgon]

Moose, just out of curiosity, is that new quad-Xeon monster in service yet? If so, I imagine this is a golden opportunity to see how it performs under a heavy workload.

 

[FR43k]

Isn't it strange that only 1.5 M extra blocks on a total of 53M+ blocks causes this problem? This is not th&aacute;t much! This is not a megaflush. If WE did a megaflush we could do it for a week and flush 6M+ every day! Thats a megaflush. This are just some blocks that members collected and which needed to be flushed (Moose ordered us!) There are more people saving blocks. People and miniteams do it to make it to the top 30 stats of the DPC or to get high in the daily overall stats at dnet. WE WERE NOT SAVING FOR A MEGAFLUSH MOOSE!!! Also your members but why not think of the japanese teams? So who's to blame? DPC is easy because they are responsable for 5% of the total output but if 1.5M blokjes extra on 55M+ total (less than 3% increase) clogs up the server I should get me new hardware because it fails. Wait lets do another thing.

Lets blame AT because of the big increase in Ta's output the last couple of weeks.

first => People should not do MF's -> Ok, DPC is to big for that.
Second => People should not save blocks -> Ok, our miniteams are to big too.
third => Dnet should get new hardware because DPC will do 5M every day within 5 months.

Good job done Moose :frown:

 

[Dutchman2000]

THERE ARE NO ORGANISED TEAM-WIDE MEGAFLUSHES.

distributed.net blaiming us with entheusiasme is just ideotic. And yes, I'm pissed. They have started a thread on our forum with factual incorrect information, after all the work and effort we put in for them. Now there is a bug in a client, to bad. We can live with that. But all work allready done with the client has to be flushed. We ask when. They say now. Now we are the problem?

 

[Bobco]

First of all: I am a DPC memeber, so my view might be biased, but I just want to give my view on what is going on right now.

OK, so an error has been made by D.net by releasing a client with a bug. Let's all work together to prevent this from having any serious consequences. The bad thing is: everybody who has used the 8012 client has produced results that will be thrown away. Too bad, but it has to be done to make sure the results of the project are correct.

I think D.net has done a pretty decent job so far of keeping the whole thing running. One of the reasons I switched from S@H to RC5 and OGR is the stability of the server. What I've seen happening now is an error and some pretty strong reactions to what has happened when everybody started flushing.

My information:
- Moose issues a warning on this forum but does not mention a date
- Somebody at DPC sees this (or the .plan)
- some people get nervous because they have been saving work that might be invalided real soon
- people start flushing to prevent loosing stats units (the work is already lost, remember?)
- there is a little conversation on IRC between Moose and DPC crewmemebers. As quoted on the DPC forum:

< sequence > we can do that but are you sure that your network can handle it? in january we had a huge backlog for a week, now the entire world will start to flush..
< Moose > why you guys are the only ones that do mega flushes
< Moose > even though we ask you not to
< Moose > just flush we will keep up
< sequence > oke..

And a lot of people start to flush... It is unforunate this happened, but this will happen as soon as people start to have a competition within the teams. To get a spot in the daily DPC top 30 some people have to save for quite some time and when all this is flushed at the same time it can amount to a large number of blocks and stubs.Anyway, I.m looking at the stats right now and the output for OGR is not that much higher than usual. Don't know about RC5 though....

 

[DSmarty]

Within DPC subteams and individuals save a little so that they are in the DPC daily stats. Usually this isn't a problem, if 10 teams flush every 10 days the daily output will be roughly the same (as you can see).
At a sudden moment dnet announces that the 466 client is buggy and that the workunits will be refused &quot;soon&quot;.
We go to their IRC channel and ask what to do, they say &quot;Flush now,&quot; we say that there are many blocks saved by inviduals and subteams, despite they say &quot;Flush now&quot;.

What the hack did we wrong ?????


Moose: I'm very sorry, but this time YOU are wrong.

 

[Dijkhuis]

< sequence > hi, could you tell me when you start refusing blocks/stubs from 8012? since we (dutch power cows) will probably have a lot of blocks in our buffers (remember january 17th) and we already backlogged dnet once and don't want to do it again and this time the entire world will flush..
< Moose > i would suggest you start flushing asap
< sequence > we can do that but are you sure that your network can handle it? in january we had a huge backlog for a week, now the entire world will start to flush..
< Moose > why you guys are the only ones that do mega flushes
< Moose > even though we ask you not to
< Moose > just flush we will keep up
< sequence > oke..