Production Dropping?

[unionpacific59]

I have not shut down a single machine, yet my output has gone from 37 Mcems down to below 30. I am getting blocks that only 70-80 tests long but they are slowing the machine down buy more than 60%. Anyone else seeing this?

Larry

[Greenbank]

The admins have decided to let the second-pass queue catch up and so no more large tests (i.e. n~10M) are being handed out for a while.

You are now being given the second-pass entries (n~2.7M) until it has caught up a small bit.

The Mcems score for these blocks is lower than that of the high n blocks.

The good news is that everyone else is in the same boat, everyone else will be seeing their speeds drop. No-one is getting an unfair advantage.

[Mystwalker]

The queue got populated with double-checking tests.
First-time testing is at 10M, second-time only at ~2.7M. Currently, the chances per time that the latter yield a prime are a lot(?) higher. The break-even point is said to be somewhere near the half of first-time testing, thus almost 5M at the moment.

The rest is already said by Greenbank, who was faster than me... :spank: :D

[unionpacific59]

Does anyone know when these second pass test will be done? I have gone from 37 kcems down to 22 kcems and I have not shut down one computer. This makes me think that I am wasting my time, computer power and $$$. I am quite sure that I am not the only one who feels this way. I am donating about $250 a month to run the crunchers here in Florida.

Larry

[royanee]

Honestly, you are actually doing the same amount of work, and in fact, when the scores are refactored under the new scoring system (not yet in place) your production during this period of time and before will be essentially equal. Do not worry, your CPU is not being lost or even wasted. It is more likely that we will find a prime during the course of this second pass catch up than find a prime in first pass over the same period of time. :)

[maddog1]

cEMs is a flawed unit, it artificially increases with the larger n's!
A more accurate unit (probably MFlops or something like that) will be implemented in the v3 client, to resolve this silly "problem"
Your performance is exactly the same regardless of k/n pair, no matter what the client indicates. Consider this a bug.
It was said by the admins of the project that sometime in the future, this problem with the lower n's will be amended and previous scores will be adjusted to reflect actual contribution to the project in terms of actual CPU time/power invested and not an inaccurate and unjust arbitrary measuring unit.
This is of special interest to older users like me, that have done lots of tests at low n's years ago, just to see all the newcomers blast them out of the top positions with their inflated scores...
Hang on with the project, noone is cheating you ATM. ;)

[[DPC]Mobster]

Originally posted by Greenbank

The good news is that everyone else is in the same boat, everyone else will be seeing their speeds drop. No-one is getting an unfair advantage. [/B] Not entirely true. Someone who ran SoB for a month in the beginning of this year, will have had considerably more output than someone who will be running it a month now.

For me it is a reason to put my computers to another project temporarily. I know that this way I will certainly be playing catchup :)

[Vato]

Well the 24 hour test completion rate has dropped from ~1800 earlier this week to ~1300 now - it looks like we've lost 30% of our crunchers. How many of them will come back? Lets go back to the first pass until we have the new scoring mechanism...

[vjs]

Well the 24 hour test completion rate has dropped from ~1800 earlier this week to ~1300 now - it looks like we've lost 30% of our crunchers. How many of them will come back? Lets go back to the first pass until we have the new scoring mechanism...

Actually this is not true what your actually seeing is the incresae in time required to complete tests.

Remember that when we started this secondpass n's were around 2.7M now we are handing out secondpass n's greater than 3.3M.

What you have to look at is the number of active IP adresses in the last 24 hours.

Normally this number hovers around 1000 anywhere from 1020-990 typically. Today when I checked it was 1065, this is actually the highest I've ever seen it.

-----------------

Production will also continue to drop as some of those slower machines turn in their n=9M's for n=3M's. The true number to watch is the active IP addresses not the number of tests.

[Joh14vers6]

Originally posted by vjs

Production will also continue to drop as some of those slower machines turn in their n=9M's for n=3M's. The true number to watch is the active IP addresses not the number of tests.
Behind one IP can be more than one computers. People could have changed there fast computers to other projects and let the slow stay on SoB.

[vjs]

The question is how many actually have multiple machines behind one IP? Also there are alot of people who simply set and forget, these people wouldn't even notice the change. It's also possible that the 24-IP is now higher b/c alot of people don't submit intermediates, this would more than compensate for a loss in CPU/people.

If you look at the graph today you can see that it's starting to level out.

We would have to really run secondpass for a couple weeks before we could actually see if we are losing people. And your right by that time it would be too late etc.

Also if the scoring were fixed, I'm sure we would actuallly gain people b/c of the fast test time.

[[DPC]Frentik]

I have multiple machines on one IP....

I download multiple tests with SBqueue from my home PC, then I write down the k/n values for each test.
At work got 3 P4's running SoB only at night, I enter each test manually in the registry, and when they finish a test i take the z12345678 file back to to my home PC to finish the last few blocks and then I send in the result.

So i think anyone that is using SBqueue shows up with only one IP but he/she might have multiple PC's running on that SBqueue....

BTW: This works great for 1st pass test which run for ~120 hours and i only need to edit the registry and copy some files once/twice a week but I cant do this for the short running 2nd pass tests this means atm i'm only running my home PC for a few hours a day. :( So for me I would like to go back to larger first pass tests ASAP

[vjs]

This is very true....

You might want to consider using P-1 in the mean time... it;s not the same as running the client but it might be alot easier for you personally in the future.

[Joh14vers6]

Originally posted by vjs
This is very true....

You might want to consider using P-1 in the mean time... it;s not the same as running the client but it might be alot easier for you personally in the future.
Scoring of P-1 is not intregated in the prp-stats. Only PRP-stats are supported yet at DPC.

[Keroberts1]

if you can finish the tests at work yo uonly have to write down the tests data and have the residue to prove its prime. I believe there is a manual way of entering these. This would eliminate the need to bring home disks too.