Well what I read on page two of the message boards, and I was guilty of it as well. The number of work units under "store up to an addiction (max of ten) days of work", this number should be set to zero all the time. At first I had it set to 10, and my reasoning being is that if it was set up by rosetta it wouldnt hurt my points. WRONG the setting is set up and managed by berkley so it drastically drastically hurts points. Work units do not complete on time, and they dont receive bonus points. I had to cancel like 200 work units when I read that. Spreading the love~~~

Well what I read on page two of the message boards, and I was guilty of it as well. The number of work units under "store up to an addiction (max of ten) days of work", this number should be set to zero all the time. At first I had it set to 10, and my reasoning being is that if it was set up by rosetta it wouldnt hurt my points. WRONG the setting is set up and managed by berkley so it drastically drastically hurts points. Work units do not complete on time, and they dont receive bonus points. I had to cancel like 200 work units when I read that. Spreading the love~~~

There aren't any bonus points. You get credit for tasks you submit. If they're late they might not show granted credit immediately but they will show credit if you look in the individual tasks, although it might take 24hrs for that credit to appear. 1-2 days is a sensible cache to allow continued running when the Rosetta servers are down.

---

well there seems to be alot of talk on these forums about bonus credit and missing deadlines. that would discredit alot of people. I would like a response from a possible moderator. Not to hate, but your the first person whos ever voiced otherwise in my humble opinion.

As far as I understand, the scoring of points DOES include the turn-around-time of the work. IE. If your machine turns around work units faster (ie. the time gap between when the work unit is downloaded and when it is finished and submitted is smaller) the points for the work unit are adjusted upwards.

This makes sense... As someone who works in data science myself, I would hate to be forced to wait for 10+ days for an answer to a query just because someone managing a server somewhere decided to keep a queue of 10 days of work.

Hey Timo, I sent you some messages when I read your posts about noise and possibly power issues holding you back from buying more computers. Im not saying you need more now that I look at your computer list, but at least youll sleep better at night. Hit me back!

Yes, a faster host tends to earn more points per hour of crunching. But no, there is no difference in points granted for the same amount of work returned 10 hours after it is issued, as compared to work returned 40 hours after it is issued.

Credit is based on the models (you see an indication of the number of models you have completed in the graphic) your machine completes. The rate of model production varies by different types of work, so the servers maintain a multiplier for each type of work issued. This is based on the running average of CPU time per model as compared to the machine's benchmark.

The whole idea is that the credit system is fair. It is based on the amount of useful work produced and returned to the project before the deadline. There are no "easy points". Work returned after the deadline is unlikely to receive any credit. Work units that complete on time, but with errors, will be granted some credit by a nightly script, which only shows when you look at the details of the task, it does not show on the task summary page.

---

Rosetta Moderator: Mod.Sense

As far as I understand, the scoring of points DOES include the turn-around-time of the work. IE. If your machine turns around work units faster (ie. the time gap between when the work unit is downloaded and when it is finished and submitted is smaller) the points for the work unit are adjusted upwards.

This makes sense... As someone who works in data science myself, I would hate to be forced to wait for 10+ days for an answer to a query just because someone managing a server somewhere decided to keep a queue of 10 days of work.

Err... what?

That's not what happens at all. Tasks are sent out with a deadline - usually 14 days. So if you return tasks a day early or 13 days early they're both received in good time.

I have no idea what makes anyone think what was said in the original post. There are no bonus points for completing tasks in "runtime +1 second" or lost for completing in "deadline -1 second".

If so many tasks are in a queue that the oldest ones fail to meet the 14-day deadline, that's a problem (but not a problem if they get returned before the next guy completes the task - that's their problem).

The only reason to keep a buffer of tasks is to ensure you have something to work on if there's an extended delay in any tasks being available at all, which is mercifully rare.

---

Well my personal reason for keeping such a big log of files is to not keep connecting to the host server. As-in people in diablo 3 used to quit games to access the auction house to much, each time creating server traffic, blizzard who makes diablo 3 eventually complained and I beleive set up safe guards, cant remember to well. Who knows if this contributed to them removing the auction house from the game. But I was always suspicious of bonus points, and its the first time I actually posted anything about it, as I always *used* to keep a cache. I gotta say over 200 rosetta tasks at a time is way to much, and maybe someone could adjust it to idk, 30 at max comparable to desktops, more for servers. Seems like a waste, because im thinking no one is ever going to count how many days of work they have. Server traffic barely ever goes down, like never ever, thats why its a server. I also gotta thank you for clearing up the bonus points, bothered me for a long time as a ex-cache keeper. But in the end, why risk someone even completing the tasks before you. I personally saw my points jump dramatically when I cleared my 200 wu cue and set to 0 days.

I tend to find myself beleivable, as the only person I know of to actually get an answer about bonus points. Yea, no ones heard that before, must have came out of left field.

Well my personal reason for keeping such a big log of files is to not keep connecting to the host server.

That's quite legitimate for a whole variety of reasons. You can regulate this for yourself if you go to OptionsComputing preferences in Boinc and select the Daily Schedules tab.

I gotta say over 200 rosetta tasks at a time is way too much, and maybe someone could adjust it to idk, 30 at max comparable to desktops, more for servers. Seems like a waste, because I'm thinking no one is ever going to count how many days of work they have.

The Computing tab allows you to select a certain number of days worth of work and calculates it for itself depending on how much you have Boinc active per day. Boinc defaults to 0.25 days fyi, but I have it set to 2 days to cover all eventualities.

Note, if you're only going to connect (say) once a day, you need to ensure you bring down enough tasks to last you until your next connection time.

Server traffic barely ever goes down, like never ever, that's why its a server.

That's true, but it has happened - often during holiday periods or weekends. There have also been occasions where tasks run out. And we've had a few episodes of a lot of new crunchers arriving (a few thousand a day for a week) which tends to wipe out the tasks available to download. Lots of unpredictable things.

I personally saw my points jump dramatically when I cleared my 200 wu cue and set to 0 days.

Points tend to get awarded within a few minutes, but you have to connect a 2nd time for your local information to know about it. If you arrange to only connect once a day, it'll take until the next day for you to see it.

Hope that helps. Basics like these are rarely spoken about, it's true.

---

i tend to be a part time cruncher hence my preference is to simply retrieve 'just enough' work (normally 1 set for all cores and another set pending) let it crunch so that the results would be returned promptly

beyond points the scientists are waiting for the results, turning that around promptly helps to shorten the queue (for everyone) and that the scientists could collate the results earlier (perhaps to plan the next dependent steps) & reduce hogging up server resources that could be freed up for other tasks :)

This "how fast you return your work" credit bonus is used by GPUGrid, but not Rosetta. GPUGrid gives you a bonus for returning their work in under 24 hours or so.

---