Posted 2220 days ago by
miw
This is a "Fair Warning" message. Have you been watching Boinc.BE and Boinc Synergy close up on you? At current figures Boinc.BE will take second place (by TC) away from you in 9-10 days. Boinc Synergy will be right in that area as well by that time. It was our hope (boinc synergy) to take 3rd by TC from Boinc.BE, but it's possible (at current rates), we'll be getting it from you. After Boinc.BE takes 2nd from you. Leaving you in 4th place, Eeegads NO. Don't let this injustice happen. Keep what's rightfully yours.
Just a heads up. In the spirit of keeping it fun.
good luck
tony
PS most BS members will be scaling down or moving to another project around the 30th. But it'll still be a close one.
Yes. We have noticed. :-) Some of us are bringing some more crunch across to defend, but since many team members are in "recovery" mode after the last AA, we might not be able to bring enough power to bear to fight off your mighty impressive assaults! If in fact you do scale back around the 30th, then, as you say, it will be close. :-)
Live long and crunch. It's all to the good!
Posted 2285 days ago by
miw
There is a lot I could say on this topic, but I'll restrict myself to this:
The benchmark system has been so thoroughly rorted that the credit in projects that still rely on boinc client benchmarks for the allocation of credit has become completely debased and I rarely crunch them anymore. QMC is a good example here. It became a total joke.
Much, although not all, of the blame for this lies with the use (Not just the creation and distribution) of boinc clients that produce ridiculously inflated benchmarks. The only reason for using these clients is to get more credit for doing less crunching. There is no ethical or reasonable justification.
Posted 2285 days ago by
miw
I gotta use it too. :-) First time any other team made a medal for us. :-)
Posted 2286 days ago by
miw
Hey - AF - thanks for the race!
We thought we might sneak in and take #1 spot for a little while, but you were too quick off the mark.
But we sure had a good run, didn't we! I had fun, and I get the feeling a lot of AF guys and gals did too.
Look out for us. We have AAs every 2 months, and you never know when we might pick a project where you are not far ahead of us.....
Congratulations on your 3M as well.... That's a lot of crunching.
Posted 2289 days ago by
miw
I didn't see that definition before, but it seems a bit useless.
Yes. It has turned out, in hindsight to be useless. Not because it is a bad definition, but because benchmarks are just open to fraud.
I reckon Astro has got it about right as to where one should be aiming.
It is impossible to get a cross-project parity that will completely satisfy everyone, partly because different projects favour different processor microarchitectures. Some are heavy on Integers and others are heavy on FP. Some are greatly affected by memory access speeds and others aren't. etc etc. 64-bit science apps are going to make this situation even more noticeable.
However, if you try to get an average credit/hr across the mix of processors that takes unoptimised SETI as the lower bound and something less than Einstein as the upper bound, then most people will think you have done a good job.
For me at the moment I find that ABC gives very similar credit/hr/core to Einstein to my Athlons, but only about 70% of Einstein to my Core 2 Duo. Haven't compared with Seti as yet. (When AA4 is over I'll collect some data for this.)
So from my point of view ABC is OK. If anything, tweak the average credit down a smidge (maybe 15%) but push the long WU factor up somewhat and reduce the threshhold where it kicks in (maybe 4 hours on an Athlon 64 2.4GHz).
BTW for my T2400 Laptop I got a 12-hr result that came in at 13.35 credits/hr/core, whereas a 3700 sec result gained 14.51, 1766 sec got 16.2, 3900 sec got 15.2, 657sec got 16.05.
Another, somewhat imprecise data point is that with the start of AA5, I switched my farm over from a mix of {CPDN, Einstein, Seti (optimised), WCG, Primegrid, RS and Rosetta} to 100% ABC and my credit/day grew by over 50%. I say imprecise because at the same time I added some more crunch and although I have tried to compensate for that in the above, it is a guess. When I go back to the mix I'll be able to quantify it better because power will stay constant for a while.
For the time being and the projects I know about, I would say the most urgent issues are in WCG (need to increase credit/hr) and Einstein (need to reduce.) ABC is in the zone but towards the top if you have Athlons.
Hendrik, I really appreciate the effort you are making to get this as close to right as possible. One day no doubt the currency of credit will become so debased and projects so many that project admins will have to find something else to "pay" crunchers to come to their projects, but people like you are pushing that evil day further into the future.
Posted 2291 days ago by
miw
Congratulations to AF for getting to 2M in this project. I think someone else has already said it, but I'll say it again, that must be pretty close to the fastest first 2M by any team on any project.
--Mark
Posted 2291 days ago by
miw
These apps run in privileged space, at least on windows boxes.
I don't think it's a big concern, on modern Windows systems you can just run BOINC as a service under a restricted user, in the same way you do under Unices.
These apps run presently only on Window boxes.
Why ? because AkosF optimizes only under Windows.
So it's a big concern.
Well, I am certainly not accusing Akosf of being a black hat (and neither is anyone else, but I say it lest anyone misinterpret.)
Just imagine somebody else out there who knows how to optimise binaries who in the game for more sinister motives. Naturally they would optimise for for Windows because it is a softer, more plentiful target.
BTW Akosf's optimisations are used in the Linux version of Einstein.
Anyway, with luck and goodwill we'll see Akosf's work show up in the project-distributed apps somewhere down the track, which would be a good outcome.
--Mark
Posted 2291 days ago by
miw
I think that is reason enough to expect people to stop using them. And then the writer of the app agreed and also suggested people should stop using it.
This is a good point, yet it's just political correctness. I really can't blame users for caring more about their energy efficiency (or credits) than political correctness, though.
All this whining about credits and optimization has already generated rather unpleasant situations in many projects, with people leaving and stopping lending their help to the crunching community.
Admins for some reason seem to be quite worried by things like cheating and cross project equalization; but cheating has a limited extension, and cross project equalization is a moot point, much like communism when you really want free market.
Free market (should) allow efficient allocation of resources, the same thing you want in crunching: thus, more credit should be awarded to efficient crunchers.
If you want to talk about economics and free markets, you might ponder this point: Project admins are in a pretty good position. They can buy computing power that previously would have cost many millions simply by installing boinc on a few decent (in some cases not so decent) server machines, installing some open source software and handing out something that costs nothing - credits! Note that I said "costs nothing" and not "is worth nothing". If I were a project admin I'd be pretty concerned about preserving the value of this currency as long as possible. It probably won't last forever.
On the science side, I share Dagorath's concerns. Validators may or may not be able to detect errors in an optimised app, depending on what the app does. The situation now is not so bad, because the vast majority of people run the project-distributed code. But if the validator is a simple comparator, then when two wrong results that are identical co-incide then they will get through. What if this WU happened to be the needle in the haystack the project was looking for, but due to some corner case the needle was missed? If this event became probable, the only way to reduce its probability is to increase quorum size, and then who's crunching efficiently? I think the people running the project have to have full confidence in / responsibility for the code that is running on the project. Having many people of varying skills out there modifying binaries militates against that ideal.
These apps run in privileged space, at least on windows boxes.
I don't think it's a big concern, on modern Windows systems you can just run BOINC as a service under a restricted user, in the same way you do under Unices.
But the fact is that most people run Windows, and most people do not run BOINC in a secure manner. If I were into capturing machines, I would not care too much whether I got *your* machine or not. I'd simply be interested in getting a lot. And then I'd disappear from the scene and BOINC would be left to carry the can.
OTOH there is only a handful of people interested in doing optimization work on the science app, and I don't think they're going to put rootkits in the apps anyway :)
Actually, there are a lot of very smart people very interested in capturing botnets, and a lot of people with plenty of money prepared to pay for the use of them. The only thing that has saved BOINC so far is that there are probably easier ways to capture large numbers of machines. If more people used BOINC, the equation would change.
Posted 2291 days ago by
miw
As long as nobody can't prove there are invalids from mod-29, the people using it are right and those whining are just... whiners.
The same kind of people with "holier-than-thou" attitude problem, they pretend they care all about science and nothing about credits, instead all they want is just make sure nobody gets more credit than them.
The message boards of every and each project are ridden with them...
I'm generally pretty quick to pick up optimised apps. But in this case, the project admin did ask people to stop using them. I think that is reason enough to expect people to stop using them. And then the writer of the app agreed and also suggested people should stop using it. Double reason. In this case, I think the people who took the app down are probably a lighter shade of grey in the hat region.
The whole issue of optimised apps really concerns me for another reason, though. These apps run in privileged space, at least on windows boxes. Wouldn't be too hard to hide a rootkit in one for a little while and just hand out the links on the forum. I do think the boinc devs need to think seriously about how to secure the apps and the BOINC server system so that only apps downloaded by BOINC can return valid results.
Posted 2292 days ago by
miw
Cool! I will analyse with interest.