stats errors [FIXED]

[Keroberts1]

whats with the stats page?

[Death]

look like it's down

[wirthi]

What stats do you mean? http://www.seventeenorbust.com/stats/teams/ is working for me.

[Guilherme]

I think Keroberts1 meant http://www.seventeenorbust.com/stats/. These stats are very strange.

[wirthi]

Oh. I see :blush:

[Joh14vers6]

It is caused by this user (http://www.seventeenorbust.com/stats/users/user.mhtml?userID=7127) with overall just 79 tests finished (1 today) and 2 pending.

Thanks to kristofvt for finding this.

[allio]

Er, how did he get more than 6P from just 70 tests? :p

[Keroberts1]

perhaps he has been working on a test he assigned himself with an extremely high N value. This test would display an incredibly high amount of cem/s (keep in mind thatthis is a imperfect unit).

He could also have submitted residues from the same tests many times.

[jjjjL]

This looks to be an accident caused by a bad report. The n value came back as n=184,549,376 so it was scored incorrectly.

The test has been rescored and all is well. I just did the fix so the main stats graph will take another few minutes to update but the users graph is already under control and the user rankings will be back in order soon as well.

I'll look into adding a sanity check to the scoring so no isolated glitches can't clobber the stats again.

Cheers,
Louie

[maddog1]

Seems to me that not all is well...:Pokes:
The same user has a huge spike between 07:00 and 10:00 at 20 Feb (which is definitely after your post!) and appears now to have 134 tests completed, up from 79 two days ago...
Either we have a major supercomputer unleashed on the project OR a completely rogue client.
My vote goes to the 2nd option :D
Please investigate...

[Matt]

Originally posted by maddog1
Seems to me that not all is well...:Pokes:
The same user has a huge spike between 07:00 and 10:00 at 20 Feb (which is definitely after your post!) and appears now to have 134 tests completed, up from 79 two days ago...
Either we have a major supercomputer unleashed on the project OR a completely rogue client.
My vote goes to the 2nd option :D
Please investigate...

I'd be interested to see a super computer with this much power:

Equivalent power (est.) 1.82 PHz

[vjs]

Well he is at it again???

Any ideas how he is doing this????

[Keroberts1]

probably editing the registry to give himself a huge N test so he has huge # of cems because he's using a huge fft size

[pcdeatherage]

Year or so ago I accidently put in the decimal number as a hex (or visa versa) and caused a relatively larger spike - perhaps he did this a second time trying to fix it...

or has discovered something that he wants the powers to be to find and fix...

or someone's just being a jerk...a note of explanation would be appropriate...

[maddog1]

Yet another huge spike.
Project stats have freaked out (check the graph...) and of course he's got #1 in all 3 rankings :rolleyes:
Shouldn't he be banned or something as a repeat offender?

[Theadalus]

What's this person using, the Earth Simulator? :crazy:

[allio]

This issue is most definitely not fixed :(

[ssvegeta1010]

I say that there either should be a ban in order, or at least some regulatory action to see whats up with these spikes.

[Joh14vers6]

Originally posted by ssvegeta1010
I say that there either should be a ban in order, or at least some regulatory action to see whats up with these spikes.

First we need to know what the cause is and then the statsserver can be patched the right way and if the cause is defenitely caused by eewee then Louie, Dave or Mike could consider a ban for eewee.

Greetings,
Arnoud

[Matt]

I noticed his "equivalent power" was in the range of Thz, but I just noticed mine is massively wrong:

"Equivalent power (est.) 39.80 GHz"

Mine is more like about 8Ghz at most.

[Stromkarl]

If eewee was using a large n value, it would show up in the secret stats page (http://www.seventeenorbust.com/secret/) here. However, it shows that first pass is no more than 9999967 and largest prime is no more than 13999982. For those who ran largest prime n values, what was the highest output? Not in the 65000G range, I am sure.

With this said, eewee has done 196 tests that do not show outside our range on the above page. However, DVNT1 (at position 2) has done 3991 tests. This means eewee has done 1/20 the tests, but has a score almost 9x higher, all without showing up outside the ranges listed on the secret stats page. Impressive! I would like to know how eewee does it. Could this be a cluster computer helping out?

Stromkarl

[Mystwalker]

Perhaps he has checked in 196 small tests (with n like 100) and one with n like 184,549,376...

[hhh]

I think if he really ran a test, he didn't get it assigned but put it in the registry on his own. Thus, it will not appear in the secret stats page, as this is an assignation stats page (right?).

Either eewee is serious, in which case he has serious problems with his machine, probably without remarking it ( :confused: but why else doesn't he find the way to the forum?). In this case, I don't know if one should make an exception of the zero-email policy before banning him; I think not; its a promise not to email in my opinion.

Or he is somebody who thinks it's funny to hack the stats page somehow, claiming like this his 15 minutes of fame. This is possible, I think, because the project attracts more and more people, so more freaks, too, I guess.

I think one has to protect the stats page from bogus returns by excluding tests with too high n. It is not worth waiting for eewee to help us. Perhaps it is possible to check if his 196 tests were serious or not.

Until the problem is fixed, one could ban him temporarily perhaps so that he remarks that there is a problem (just in case he is serious).

[Matt]

ROFL I just realised I recognised the username of this person and I told them about the project and told them to set it up, at least I think it's him. He runs FreeBSD and I very much doubt any of this is in any way deliberate. I'll email him and give him a call when he gets out of college. Please don't ban his account any time soon!

[Keroberts1]

the simplest thing ot do would not give any credit for any tests not assigned by the server (or tests that have been expired by the server). This would eliminate the problems entirely.

[Matt]

I've just spoken to him on the phone, he's gonna sort it out when he gets home. He was completely unaware of the problem.

[Matt]

Ok, seem's all the communication is going to be through me. He's now shut down the 1 client that was running on his computer, stats are:

AMD Athlon XP3200+
1GB PC3200 RAM
FreeBSD 5.3
SoB Client 2.2

He's emailed me his complete log from some time in 2004 onwards, I've uploaded it to my server here (http://mattford.org.uk/sclient.log). I've also got the current cache file and everything so if an admin wants it, PM me their email address and I'll attatch it to them.

There were only 2 work units associated with his account which both appear to be coming from his home IP address, nothing seems particularly odd about them.

Generally it's quite a mystery, no idea what's gone on, but hopefully it's all stopped now.

[maddog1]

A very interesting log indeed...
I found the problem here:
[Wed Feb 16 15:43:20 2005] iteration: 8090000/8107315 (99.79%) k = 10223 n = 8107301
[Wed Feb 16 15:46:28 2005] resolving hostname
[Wed Feb 16 15:46:28 2005] opening connection
[Wed Feb 16 15:46:28 2005] server busy -- block added to submit queue
[Wed Feb 16 15:51:01 2005] resolving hostname
[Wed Feb 16 15:51:01 2005] opening connection
[Wed Feb 16 15:51:01 2005] server busy -- block added to submit queue
[Wed Feb 16 15:54:48 2005] iteration: 8100000/8107315 (99.91%) k = 10223 n = 8107301
[Wed Feb 16 15:55:10 2005] resolving hostname
[Wed Feb 16 15:55:10 2005] opening connection
[Wed Feb 16 15:55:10 2005] server busy -- block added to submit queue
[Wed Feb 16 16:01:01 2005] resolving hostname
[Wed Feb 16 16:01:01 2005] opening connection
[Wed Feb 16 16:01:02 2005] logging into server
[Wed Feb 16 16:01:03 2005] login successful
[Wed Feb 16 16:01:14 2005] n.high = 8104193 . 1 blocks left in test
[Wed Feb 16 16:08:06 2005] residue: 13730455BC3ABB39
[Wed Feb 16 16:08:06 2005] completed proth test(k=0, n=184549376): result 3
[Wed Feb 16 16:08:06 2005] connecting to server
[Wed Feb 16 16:08:06 2005] couldn't report to server, retrying in 1200 secs
[Wed Feb 16 16:28:06 2005] connecting to server
[Wed Feb 16 16:28:06 2005] couldn't report to server, retrying in 1200 secs
[Wed Feb 16 16:48:06 2005] connecting to server
[Wed Feb 16 16:48:06 2005] couldn't report to server, retrying in 1200 secs
[Wed Feb 16 17:08:06 2005] connecting to server
[Wed Feb 16 17:08:06 2005] couldn't report to server, retrying in 1200 secs
[Wed Feb 16 17:28:06 2005] connecting to server
[Wed Feb 16 17:28:07 2005] logging into server
[Wed Feb 16 17:28:07 2005] requesting a block
[Wed Feb 16 17:28:08 2005] got proth test from server (k=67607, n=8307251)

I have no idea what could have happened here, but this is the number Louie mentioned earlier in this thread. Obviously a runaway client then...
A half-careful scan of the log showed no other obvious faults, be my guest to scan it more carefully (I got dizzy after a while...) :D

[maddog1]

Some more detective work (this can be pure speculation, but numbers match up closely enough, so I mention it-a nice theory anyway...) :)
1)John14vers6 posted at 3:33 PM on 18/2 that the user had 79 tests complete.
2)I posted at 11:51 PM on 20/2 that he had 134. (times for all the forum posts are GMT+2, since I am Greek)
DIFF between 1) and 2) is 55 "tests"
3)Stromkarl posted at 2:48 AM on 24/2 that he had 196.
DIFF between 2) and 3) is 62 "tests"
4)Now he shows at 202.
DIFF is 6 "tests"
TOTAL DIFF is 202-79=123 "tests"
I assume times in the log are GMT, since Matt tells us he knows the guy and he's UK based himself.
If you count the times the client contacted the server between 18/2 1:33 PM and 20/2 9:51 PM, it adds up to 59 times (check the log)
If you count the times the client contacted the server between 20/2 9:51 PM and 24/2 0:48 AM, it adds up to 65 times
If you count the times the client contacted the server between 24/2 0:48 AM and the end of the log, it adds up to 6 times.
Total times the server was contacted SUCCESSFULLY (n.high= lines...) in roughly the same time span were 130.
Small differences in the numbers may be attributed to PC clocks going a few minutes wrong, the time between the user saw the number of "tests" and posting about it in the forum etc etc-The last space which conveniently started during the night at 0:48 AM when the client wasn't working to make reports matches perfectly and the others are really close.

MY THEORY: Each time the client was contacting the server, somehow it was reporting the huge test found in the logs as "complete" again and got credit again for it. Probably it all sums up to a corrupt cache issue, maybe someone with access to the server side logs or knowledge about the client's inner workings can improve my gibberish :D
PS: I seriously hope I got the timespans correctly, GMT, EST blah blah blah crap...dizziness from looking at the log doesn't help at all ;)

[Matt]

That seems to make sense to me, I think some server side checking is probably nessecary to weed out things like this... Hopefully it should be easy to find the reported work since we now know it is for k=0 and remove it. I'm guessing some kind of maths based on the K number is used to calculate the speed and we ended up with infinity rounded down to the highest number that would fit in the allocated space or something? Anyhew, it seems that we've got to the root of the problem.

[garo]

It may also make sense to get one of the project managers to run the originally assigned test on an AMD to make sure it is not a software error.

[Matt]

This error is marked as fixed but the erroneous stats still exist in eewee's name, should they not be removed?

[ssvegeta1010]

eewee's stats are now also normal.
This may finally be resolved. (Of course after his erroneous tests have been reassigned)