Yet again, Dr. Andersons only reply to Eric about credit was:


David Anderson to Eric, BOINC
More options Mar 5

Given the goals of
1) minimizing variance of claimed credit for a given result, and
2) parity between projects (on average)
it seems like the most practical solution is to use FLOP counts,
but scale them as needed for 2).


-- David

- Show quoted text -

and to show you Eric did think this through, here's the question:

Eric J Korpela to BOINC
More options Mar 4

SETI@home enhanced is coming up on release and it's getting to the
point where I need to decide what to do about credit calculation. The
big problem is the disparity between calculating credit based upon CPU
time and calculating credit based upon work actually performed (i.e.
FLOPS).

If you go to boincstats.com and look at the total recent credit you'll
see something that looks like 2000 TFLOP/s. That number is about a
factor of 10 higher than the total number of FLOPS actually begin
done. It's actually the theoretical maximum performance of BOINC
rather than an actual performance. (And people writing papers about
BOINC should be aware of that lest they provide incorrect
information.)

If I calculate the actual number of FLOPS performed by S@H enhanced
and compare it to the run times, the average ratio between theoretical
(i.e. the benchmarks) and actual performance is about 4. The ratio is
strongly dependent upon processor speed with 2 GFLOP processors having
a larger ratio than a 1 GFLOP processor. That's not really surprising
since a memory bandwidth limited process should scale with the FSB
speed, not with the processor speed. (See
http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=111 for some
plots)

So the dilemma I face is how to grant credit in a way that's fair.
Within a project, granting by "work done" is OK, but when we're
talking about multiple projects it gets more complicated. I've
already got people predicting a mass exodus from SETI@home when people
with "fast" (high benchmarking) machines realize they will get more
"credits" from projects that grant by CPU time rather than by work
done.

In the S@H beta project, I've temporarily resolved the problem by
granting 9X the number of FLOPs. Yet I still get complaints from
people whose machine have high benchmark scores. (As can be seen from
many other threads in the beta forums).

If granting by actual work becomes a problem, the two resolutions that
I can see are to either go back to CPU time for credit and discard the
idea of measuring work done, or to modify how BOINC calculates scores
when using the CPU time method, so as to more accurately show
performance.

One possibility would be to change the cpu_time derived credit to be:

credit=cpu_time/(4/memory_bandwidth+1/FLOPS+1/IOPS)*constant

If any of the number (bandwidth/4, flops, iops) is significantly lower
than the others, it would then become the dominant term in the
calculation... I suppose it would be possible to use the memory
bandwidth in combination with cache size/bandwidth and working set
size to come up with a better measurement.

Other ideas?

Eric

Daniel

Right now many are saying things that go outside what we would normally do/say... I am also guilty...

Yes tommorows outage as long as UCB Staff remembers to to turn everything "on" will be interesting...

Ahm...The weekly outage is not until TOMORROW. Sorry, Al...

I will wait to make my final decision until there has been some kind of response from project management regarding this fiasco. Then, we'll see what happens.

Regards, Daniel.

I suspect that they are hoping it blows over... When everyone is done getting things off their chest then things go back to normal. Users have problems, Users help Users...

Regards
Al

---

Please consider a Donation to the Seti Project.

@Jack Gulley,

Just out of interest why do you need the return immediately option?

And please don't say its because you run old hardware, I run a p2 300 on Beta.

Andy

@Jack Gulley,

Just out of interest why do you need the return immediately option?

And please don't say its because you run old hardware, I run a p2 300 on Beta.

Andy

It's extremely useful, you can spot failed hosts immediately because they fall back in your hosts list.

Regards Hans

P.S: With seti enhanced, no answer in 12hrs just means you got yourself a really tough WU :o)

---

Why not use a number that kept the granted credit at or very near the current credit per hour level rather than reducing it?

As has been posted many times before- they did.
But it was based on the official client, not the optimised ones.

---

Grant
Darwin NT

Ahm...The weekly outage is not until TOMORROW. Sorry, Al...
Regards, Daniel.

Ho, hum.....

2006-05-16 06:04:57 [SETI@home] Started upload of 22fe99aa.24201.1074.436076.3.173_3_0
2006-05-16 06:05:00 [SETI@home] Temporarily failed upload of 22fe99aa.24201.1074.436076.3.173_3_0: error 403
2006-05-16 06:05:00 [SETI@home] Backing off 3 minutes and 41 seconds on upload of file 22fe99aa.24201.1074.436076.3.173_3_0
2006-05-16 06:08:42 [SETI@home] Started upload of 22fe99aa.24201.1074.436076.3.173_3_0
2006-05-16 06:08:44 [SETI@home] Temporarily failed upload of 22fe99aa.24201.1074.436076.3.173_3_0: error 403
2006-05-16 06:08:44 [SETI@home] Backing off 7 minutes and 19 seconds on upload of file 22fe99aa.24201.1074.436076.3.173_3_0
2006-05-16 06:14:11 [SETI@home] Started upload of 22fe99aa.24201.1074.436076.3.177_1_0
2006-05-16 06:14:15 [SETI@home] Temporarily failed upload of 22fe99aa.24201.1074.436076.3.177_1_0: error 403
2006-05-16 06:14:15 [SETI@home] Backing off 29 minutes and 31 seconds on upload of file 22fe99aa.24201.1074.436076.3.177_1_0
2006-05-16 06:16:04 [SETI@home] Started upload of 22fe99aa.24201.1074.436076.3.173_3_0
2006-05-16 06:16:07 [SETI@home] Temporarily failed upload of 22fe99aa.24201.1074.436076.3.173_3_0: error 403
2006-05-16 06:16:07 [SETI@home] Backing off 17 minutes and 5 seconds on upload of file 22fe99aa.24201.1074.436076.3.173_3_0

---

@Jack Gulley,

Just out of interest why do you need the return immediately option?

And please don't say its because you run old hardware, I run a p2 300 on Beta.

Andy

It's extremely useful, you can spot failed hosts immediately because they fall back in your hosts list.

Regards Hans

P.S: With seti enhanced, no answer in 12hrs just means you got yourself a really tough WU :o)

But, even on that computer my results have returned before others crunching the same unit, so yes, it is returned ok but I don't conclusively know until it has been validated. And except for a couple of timer errors, which occured before it was given different hard disk and clean install, I trust it enough by just checking ocassionally that it is crunching and not stalled. Actually had more problems, on Beta with my Pent M but thats probably because it has crunched many more units.

Andy

Ahm...The weekly outage is not until TOMORROW. Sorry, Al...
Regards, Daniel.

Ho, hum.....

2006-05-16 06:04:57 [SETI@home] Started upload of 22fe99aa.24201.1074.436076.3.173_3_0
2006-05-16 06:05:00 [SETI@home] Temporarily failed upload of 22fe99aa.24201.1074.436076.3.173_3_0: error 403
...

Might be worth checking things at your end.
No problems here, Server traffic looks Ok & status page is still green.

---

Grant
Darwin NT

I can still post, so at least my internet connection is working...
What's the name of the upload host again?

Regards Hans

---

What's the name of the upload host again?

setiboinc.ssl.berkeley.edu is the scheduler. Don't know about the upload/download server(s).

---

Grant
Darwin NT

Hmmm... Strange

POST /sah_cgi/file_upload_handler HTTP/1.1

User-Agent: BOINC client (i686-pc-linux-gnu 5.3.12)

Host: setiboincdata.ssl.berkeley.edu

Accept: */*

Content-Type: application/x-www-form-urlencoded

Content-Length: 288

Expect: 100-continue



HTTP/1.1 403 Forbidden

Date: Tue, 16 May 2006 04:42:29 GMT

Server: Apache/1.3.33 (Unix) mod_fastcgi/2.4.2

Connection: close

Transfer-Encoding: chunked

Content-Type: text/html; charset=iso-8859-1



138

<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML 2.0//EN">
<HTML><HEAD>
<TITLE>403 Forbidden</TITLE>
</HEAD><BODY>
<H1>Forbidden</H1>
You don't have permission to access /sah_cgi/file_upload_handler
on this server.<P>
<HR>
<ADDRESS>Apache/1.3.33 Server at setiboincdata.ssl.berkeley.edu Port 80</ADDRESS>
</BODY></HTML>


0

---

Just out of interest why do you need the return immediately option?

I like to keep track of my machines as they also function as low access rate data servers for a process that runs on one of my older machines, and I know (or did know) how long it takes each one of them to complete a WU. But I do not always have direct access to them or the local network. I do however almost always have access to the Internet and the Berkeley servers, and can check there on one publicly accessible web page, day or night, to see when the last contact with each of the systems was. This allows me to remotely check on them and detect system hangs and power glitches that get through each machines UPS. (This is South Florida after all.) I then know when my attention at home is needed and how important the outage is.

With Seti@home Classic and SetiQue I could do the same by checking the total count progress as I knew my average rate of production, but not check individual machines. One reason I stuck with it to the end. A neighbor who remotely maintains a large number of his clients systems scattered across the country was doing the same thing by setting up a Seti@home Classic account for each account or small group of systems. This way he could check on them during times of the day when he did not normally have (for security reasons) direct access to them. Security issues with things like programs updates being automatically downloadable through company firewalls has ended his use of Setiathome except on his own personal systems.

My original interest in optimized applications was so I could detect much sooner when a system of mine was down, and why I don't care for longer running Enhanced WU's. (I have no interest in any of the other BOINC based projects!) Interest in completion came later as part of The Planetary Society team. The delaying of reporting of WU's by several hours or much more has caused numerous wasted trips home. Trux's optimized client solved that problem. But of course I will soon be back to five or more hours before a system outage is detected. And will most likely have to find some other way to check on them when I am not at home (or stay home more). Setting up my own Internet server is not an option allowed or supported by my ISP, and I really don't care for another process running on each of my systems connecting to the Internet or that reports to some commercial web site. SetiQue was nice in that I could isolate all Internet contact to one system and not expose the others.

@Jack Gulley,

I suppose in you case it makes a little bit of sense, guess I'm just not that competative these days.

Must be annoying to live in a supposedly modern country, and get such a bad electricty supply. But then again looking at the past history of railway track, freeways, cars etc it is probably what I should expect in your country ;-)

Andy

Must be annoying to live in a supposedly modern country, and get such a bad electric supply. But then again looking at the past history of railway track, freeways, cars etc it is probably what I should expect in your country

The electric supply is quite good here, now that five hurricanes in two years took down the bad poles and stripped all the trees down so they are no where near the power lines. Interesting to watch what 120 mph wind driven water can do to a tree. Stripped all the bark off of one 24 year old Ficus tree in the front yard as I watched and left only the limbs in the upper 20 feet of a 115 foot tall formosa pine tree. But that only makes all the lightning strikes from the afternoon thunderstorms more likely to hit the power lines near you and cause glitches and brief outages. Most people are either not home or know to leave their computers off that time of the day. Seti crunchers know to have a small UPS as part of their system. Seti systems crunched through it all until the last hurricane took down the main power lines coming out of the power plants. Sure was dark at night for a week.

When hurricanes take out Internet access for several weeks, I could always pack up the SetiQue system and drive to a location with working phone lines to get my small farm back crunching. With BOINC, I would have to move each system. Well, not as often now with the longer deadlines.

There is usually a good reason for everything. Looks like the real reason for our current complaint about Seti Enhanced Credit is that the developers did their academic calculations on what a WU credit should be and failed to take into account how hardware really works and how inefficient their old program was, giving the Seti 4.18 over inflated credits. And they are now trying to fix that, much to everyone's dismay.

Must be annoying to live in a supposedly modern country, and get such a bad electric supply. But then again looking at the past history of railway track, freeways, cars etc it is probably what I should expect in your country

The electric supply is quite good here, now that five hurricanes in two years took down the bad poles and stripped all the trees down so they are no where near the power lines. Interesting to watch what 120 mph wind driven water can do to a tree. Stripped all the bark off of one 24 year old Ficus tree in the front yard as I watched and left only the limbs in the upper 20 feet of a 115 foot tall formosa pine tree. But that only makes all the lightning strikes from the afternoon thunderstorms more likely to hit the power lines near you and cause glitches and brief outages. Most people are either not home or know to leave their computers off that time of the day. Seti crunchers know to have a small UPS as part of their system. Seti systems crunched through it all until the last hurricane took down the main power lines coming out of the power plants. Sure was dark at night for a week.

When hurricanes take out Internet access for several weeks, I could always pack up the SetiQue system and drive to a location with working phone lines to get my small farm back crunching. With BOINC, I would have to move each system. Well, not as often now with the longer deadlines.

There is usually a good reason for everything. Looks like the real reason for our current complaint about Seti Enhanced Credit is that the developers did their academic calculations on what a WU credit should be and failed to take into account how hardware really works and how inefficient their old program was, giving the Seti 4.18 over inflated credits. And they are now trying to fix that, much to everyone's dismay.

Lived in the Shetland Isles for nearly four years, they get 60mph winds 100 days a year and when the radar staion at the northern tip gat the highest recording a wind spead measureing equipment from the USA, in the first storm it went full scale and stuck there, whoops. The lighthouse there Muckle Fluga (hope I spelt that right) has an external door 150 ft abobe sea level and it is frequently bashed in by the waves. There very few trees on Shetland they cannot surveive.

The credits for enhanced are in line(approx) with 4.18 un-optimised, and if you can speed it up they will still grant the same not decrease with speed up as was/is the case with 4.18 optimised. If you use client that does fpop counting.

Andy

I dont understand the hole thing.
All will sort out by itself.

Mike

---

With each crime and every kindness we birth our future.

I suspect that they are hoping it blows over...

The point that needs to be realized is, that there are many different motivations for participants in these projects. Some are here strictly for the "science", and couldn't care LESS about credit, and they don't understand why it is granted in the first place. At the other end of the spectrum, there are people who are here solely for credit alone. I am here for both. And BOTH ends of the spectrum, and everyone in between, are ALL doing valid "science". If they weren't, their work would not be validated. But, if you take away the motivation for doing the science from a very large subset of crunchers, then they will either go somewhere else for their credits, or they will become totally disinterested in distributed computing altogether, and shut their computers off. Either way, it will end up ruining it for everyone. It does not matter to the science at all, if 30-60 credits are granted or if 160-230 credits are granted, as they were in Seti Beta. But it DOES hurt the science if a very large number of hosts are lost from the project. And it doesn't hurt any of the participants who don't care about credits, if more credit is granted. They don't care about them anyway. So, I just can't figure out why everyone who is supposedly here "solely for the science" is so opposed to granting more credit. If they don't care about credit, then fine. But don't insult and chase away those that DO care about credit.

People with an opposing view of mine should read Ayn Rand's book Atlas Shrugged. This is an EXCELLENT example of what happens when a very large subset of people depend upon a very small subset of people with greater ability and resources. There is nothing bad about people who have less ability and/or resources, but think about how little work would have actually been done for this project if it were not for people with very large farms of computers. One person with 50 computers can do the work of 50 people with ONE computer, and vice versa. But the thing to understand is that there would not BE people with even a dozen or more computers, if it were not for credit and competition. Alienate that subset of users, and push them away, and it would take many more YEARS to do the same amount of work. Sure, there would always be a few people left, but the project as a whole would greatly suffer without the top participants. This issue is not going to just blow over, regardless of how much anyone wishes it would.

Regards, Daniel.

Daniel,
It will matter to the projects, and to BOINC, if the level of credits/time for each project is not kept approximately the same. If not people who have no allegance to a project or are pure credit hogs, will switch from project to project, credit whore's or mercenaries in effect.

If the credits are kept approx. equal then the choice is open, knowing that it doesn't matter from a credits POV which project(s) they crunch for.

As most of the projects do not reveal their source code then unless they have someone like Akosf, over on Einstein, who can reverse engineer the code and is profficient in assembler, then the chances are most of the optimisation is going to be done here on Seti.

So the choice at the moment for most users, even if they do not like enhanced, is take a limited, probably less than a month, look over on Einstein and then come back here and use crunch3r's latest optimised app or ANOthers, if that appears and is faster.

I just cannot see any other options for the credit hogs, unless they work out which project/hardware and OS combunation gives them most bang for their buck.

Andy

Just out of interest why do you need the return immediately option?

I like to keep track of my machines as they also function as low access rate data servers for a process that runs on one of my older machines...
[snip]
This allows me to remotely check on them and detect system hangs and power glitches that get through each machines UPS. (This is South Florida after all.) I then know when my attention at home is needed and how important the outage is.

there are much easier and better ways of monitoring, you could set up a dynamic DNS account, so you can always access your site remotely, and run something like Nagios on a web server (prehaps on a random port, requiring authentication, so your ISP is happy), that'll give you a lot more detail, you could also set up secure remote access, so you can log into machines remotely, and use them over the internet, to reboot them or whatever

Setting up my own Internet server is not an option allowed or supported by my ISP, and I really don't care for another process running on each of my systems connecting to the Internet or that reports to some commercial web site. SetiQue was nice in that I could isolate all Internet contact to one system and not expose the others.

Nagios would allow you to do exactly that, and it puts virtually no additional load on PCs/servers, it's used by many companies for comercial/professional monitoring of servers

it's not really fair to use the SETI system in this way, putting additional load on the servers when there are better methods available

---

Want to search the BOINC Wiki, BOINCstats, or various BOINC forums from within firefox? Try the BOINC related Firefox Search Engines