Message boards :
Number crunching :
Average Credit Decreasing?
Message board moderation
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 32 · Next
Author | Message |
---|---|
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
Well, group W is where they put us all in the case of not adhering to the norm. Ah yes, the Group W bench. Funnily enough, I was introducing that song to a colleague a few weeks ago, in the context of a different credit crisis - the global economic one. That was more in the context of suggesting that collective action was more effective than solo efforts like hers. You know, if one person, just one person, does it, they may think he's Really sick and they won't take him. And if two people do it, in harmony, they may think they're both faggots and They won't take either of them. And if three people do it! Can you imagine three people walkin' in, singin' A bar of "Alice's Restaurant" and walkin' out? They may think it's an Organization! And can you imagine fifty people a day? I said FIFTY people a day . . . Walkin' in, singin' a bar of "Alice's Restaurant" and walkin' out? Friends, They may think it's a MOVEMENT, and that's what it is: THE ALICE'S RESTAURANT ANTI-MASSACREE MOVEMENT! . . . and all you gotta do to join is to Sing it the next time it comes around on the guitar. Now, I'm not advocating that fifty people a day write to David Anderson about CreditNew - but a more collectivist approach might be needed in the end. If fifty BOINC project administrators wrote to DA, he might take some notice. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Funnily enough, I was introducing that song to a colleague a few weeks ago The short version, or all 18 minutes worth of joy? Litterbug. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Happy Yuri's Night #55, the 55th anniversary of Yuri Gagarin's historic first space flight on April 12, 1961. I'm guessing we are now at the point of Yuri's decent, when he was told "I hope you enjoyed the ride, now you can start screaming as your flaming arse plummets back to earth!" |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
... I will gladly raise the credit issue with him too, but we have to avoid antagonising him and driving him away again. Taking a small proposal for tuning the apparently important (to many) credit might help, at least until more rigorous rework of the estimate mechanism can be done. Have been trying to boil down to the simplest fix since we experimented on Albert. Usually I'm against bandaid/knee-jerk patching, but I think this is one scenario where it might help cool things down. Here I use the S.O.A.R format, i.e. situation, objective, action, result, so as to feed the collective process. Situation: Estimates (and so credit) are normalising 'peak flop counts' to applications that use dynamic dispatch (i.e. various levels and types of SIMD vectorisation), while claiming using a non-vectorised benchmark (in the cases of Windows, Mac and Linux on x86 & x64) for peak flops measure. This incorrect claim yields numbers that indicate > 100% computational efficiency, when on a heavily vectorised application on equipped hosts, driving the payout for all other applications to less than intended by the mechanism. Other instabilities exist (of varying importance) though the credit award is one where a coarse correction would be highly appreciated, even if the estrimates and stability under change are never addressed. The ratio of this scaling error can be confirmed, by comparing boinc whetstone to suitable SiSoft Whetstone, single threaded SIMD ( SSE or AVX), for any particular host. Objective: Coarse correction of the scaling error, rather than addressing any of the other more complex engineering problems to do with estimates/credits, so as to take the bulk of the hysteria/angst out of the problem. Action: Using the information already available server side, i.e. processor features, client platform, and known application feature support, devise and apply a multiplier for the claimed peak flop count, when the client only has FPU/Scalar Whetstone (i.e. Windows, Linux, Mac ) e.g. pfc_multiplier = 1 + logbase2(min( maxFpVectorLengthOfhost(), maxFpVectorLengthofApp() )); For Android, where boinc Whetstone is vectorised on later clients, but applications (known) may not use that vectorisation, assume the max vector length anyway (throughput is low at this time) For anon platform, just assume the app on the host is or will use the maximum SIMD Vector length, but then anon platform gets normalised to the most efficent claiming stock app anyway (invariably AVX enabled stock CPU app claiming >100% efficiency for our purposes). Result: peak flop counts become 'more sane', and at least roughly compensating for different technologies in the apps+hosts (which it doesn't now). Further stabilisation could be easily employed once, through the visible credits, throughputs look more within the bounds of reality (i.e. credit would still be over the place, but at least not being driven into the ground by optimisation of the stock CPU app) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
@Richard: I think there are more than you might expect, but It's very hard to be sure - we asked, and tried to answer, the same question last time we went round this circle, two or three years ago. I think there are probably more projects than you think, but they'll be the small ones, the one-man-band projects, with CPU applications only. Don't forget, CreditNew has been the default BOINC server code for nearly six years now, and the vast majority of project administrators will just take the stock code and run it. They, after all, really are "in it for the science": they don't have the skills, and don't want to learn the skills, to implement anything different. Actually, CreditNew doesn't work too badly for a simple CPU project with a single application which doesn't change very often. Rapid change trips it up badly - more on the runtime estimate side of things than the credit side. GPU applications are more problematic, and CreditNew is effectively useless for projects which only use GPUs, and don't have a CPU version to tie it back to the original definition of a cobblestone. At least, that's my understanding of the conclusions which came out of Jason's (and companions) ramble through the code jungle in 2013, and an information-gathering exercise at Albert the following year. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
The short version, or all 18 minutes worth of joy? I was explaining collective action over the telephone to a lady well into her seventies, so I kept it short.... But I did suggest she watched the film. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
At least, that's my understanding of the conclusions which came out of Jason's (and companions) ramble through the code jungle in 2013, and an information-gathering exercise at Albert the following year. Boils it down pretty well. Only one app, and you get around the cobblestone scale credits per the project supplied estimate, which is fine, since that app becomes the normalisation reference. Indeed doesn't help estimates if/when something changes though, or a project sets the #operations incorrectly. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
rob smith Send message Joined: 7 Mar 03 Posts: 22526 Credit: 416,307,556 RAC: 380 |
Both Jason and Richard forgot one of the sneaky little bits - CreditNew works OK when the voting processors are of similar performance, but if you have processors with different performance one bit of the algorithm works rather badly. When calculating the credit to be awarded the process takes the lowest Cobblestone value to be the "correct" one. Additionally the "credit multiplier", which is meant to correct for processor performance, does not work correctly, thus compounding the issue. Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
I think the "credit multiplier", and the associated flopcounter, which combine to generate the <fpops_cumulative> reported to the server, were components of The second credit system - the one which was superseded by CreditNew. And the main point of CreditNew was that it would be fully automatic - projects wouldn't need to have an Eric to make it work. Not every project has an Eric (although Eric Driver and Eric Macintosh might quibble with me saying that). Eric (our Eric) deliberately left those elements in place - they are under the control of the project. But SETI gets zero public funding, and (at the time) BOINC was NSF funded as a demonstration/experimental project for distributed computing. It was politic for SETI to use the NSF-funded infrastructure, because otherwise it would have had no infrastructure at all. Now, the boot is on the other foot. BOINC has demonstrated the success of its experiment to the satisfaction of the NSF, and the infrastructure is permanent. This has resulted in the removal of the public development funding, but as yet BOINC hasn't transitioned to main-budget maintenance resourcing. Nobody's filling in the potholes, just like in the real world. SETI, on the other hand, now has a substantial windfall of private funding. I know a little bit about good practice in the philanthropic field. In the UK, the best charitable trusts (including the National Lottery, but unfortunately not the BBC Children in Need appeal), insist that the projects they fund cost in a realistic sum for administration and infrastructure: unless the application includes that overhead, they won't fund it. Now, I have no idea of the terms of Yuri Milner's gift: whether he asked the right questions, or whether Berkeley deployed a wise negotiator. But, handled carefully, we might have some leverage to petition Breakthrough for money to repair the cracks in BOINC. I hinted as much in my email to David. But if that fails SETI might, indeed, be free to wind the clock back six years to the second credit system. |
Mike Send message Joined: 17 Feb 01 Posts: 34378 Credit: 79,922,639 RAC: 80 |
Maybe credit new is based on the european base interest rate. LOL So long it doesn`t get negative ............ With each crime and every kindness we birth our future. |
rob smith Send message Joined: 7 Mar 03 Posts: 22526 Credit: 416,307,556 RAC: 380 |
Mike - I think you might have found one of the missing links ;-) Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
kittyman Send message Joined: 9 Jul 00 Posts: 51478 Credit: 1,018,363,574 RAC: 1,004 |
Dr. Anderson USUSALLY only responds when you can prove that one plus one does NOT any longer equal two plus or minus ten digits either way. The outcome creditwise usually does not concern him much because credits are not a scientific matter, and are naturally not part of his response curve. So somebody complaining that credits....HIS credit system....are broken naturally just slide off of his back into nowhere land. It has to take a LANDSLIDE of massively wrong (or correct, in this case) information to sway the man one way or another. We all have to continue to speak up for justice in this matter until the ship is righted. Meow. "Time is simply the mechanism that keeps everything from happening all at once." |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13854 Credit: 208,696,464 RAC: 304 |
My RAC had been declining for a while, but now it's turned in to free fall. Grant Darwin NT |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
LISA Pathfinder is in free fall at Lagrange point 1 Tullio |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Dr. Anderson USUSALLY only responds when you can prove that one plus one does NOT any longer equal two plus or minus ten digits either way. One issue that software faces is how people perceive how it should work. Even if CreditNew is working 100% as designed it is perceived as broken. As credit is a means to determine performance, or contribution, to a project it seems counter intuitive to me for that to be highly variable. If I configure a machine to run a project a fixed amount of time a day. It would be 1 hour a day or 24/7. It makes logical sense that it should be contributing a fairly constant amount of processing power and by extension generate a fairly constant credit value. It make it hard to figure out what is going on when the reference isn't a consistent value. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
AMDave Send message Joined: 9 Mar 01 Posts: 234 Credit: 11,671,730 RAC: 0 |
"Average Credit Decreasing?" Is this resulting from GBT WUs or something else? I take RAC with a grain of salt, however, I employ it as a pseudo-barometer. For example, for the past two weeks, my RAC has decreased ~8.2%. Note -- For comparison in the past two weeks, on another project I crunch for, my RAC has increased ~1.5%. I've only begun receiving GBT WUs two days ago. Although these WUs require less time to complete (when they actually complete) than Arecibo WUs, I've calculated that they both receive 'roughly' the same credit/unit of time. So, I'm a bit perplexed. The 'roughly' similar credit tells me that it isn't software related, yet I've done nothing on the hardware end.[/i] |
bluestar Send message Joined: 5 Sep 12 Posts: 7264 Credit: 2,084,789 RAC: 3 |
Anyway. http://setiathome.berkeley.edu/workunit.php?wuid=2136322667 Here is an example I found when it comes to credit only. Is this because of the run time length only? Definitely something is wrong here. Hey DA! Why not listen to the users at times? They are those who are supposed to be doing the work. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
On the upside, I feel like I have 980 Titans now with how many files I am going though :) |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
On the upside, I feel like I have 980 Titans now with how many files I am going though :) Always upsides: If those triplets are legit, then we probably found a squillion pulsars, some noisy ET's, a bunch of pidgeon poop on some telescope, or some other thing nobody knows what it is yet. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
I had a good laugh at "pidgin poop" , our 600 watt Ku band transmitter was great at cooking pidgins in the winter ... I guess they liked to keep warm, outcome was not so great. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.