GPU Upgrade Shows No Improvement in Work Unit Completion

Gamboleer
Gamboleer
Joined: 5 Dec 10
Posts: 173
Credit: 168389195
RAC: 0

RE: I noticed that I've

Quote:
I noticed that I've just had a 4M day (so it says) - so yes, I'm a bit over 3M

Gary is going for the headline, MAN DISCOVERS 47 PULSARS IN ONE DAY

.

Florida Rancher
Florida Rancher
Joined: 4 Oct 13
Posts: 31
Credit: 23998436
RAC: 0

Gamboleer: Where do find

Gamboleer:

Where do find information on what the project is accomplishing? Where did you find the pulsar stats?

Phil

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109378039538
RAC: 35985209

RE: Where do find

Quote:
Where do find information on what the project is accomplishing?


On the front page of the website down the left hand column. Take a look at various pulsar search related links under the sub-heading, "Science information and progress reports".

In particular, if you look at the three links under the bullet point "Overview of Einstein@Home Radio Pulsar (Re-)Discoveries:", you will find lists of all discoveries/re-discoveries together with the names of the computer owners who made each 'find'.

Cheers,
Gary.

Florida Rancher
Florida Rancher
Joined: 4 Oct 13
Posts: 31
Credit: 23998436
RAC: 0

Gary, Thanks for directing

Gary,

Thanks for directing me to where I can see how our work is paying off. Is there a way to search by user because I'd like to see your stats and check mine from time to time.

I've always been interested in science and that's why I'm drawn to Einstein@Home. It was probably about 15 years ago that I did Seti crunching. I can't even remember the account name so my "Joined" date is off.

I've noticed that in the past 5 days my "Validation Pending" has tripled from the low 20s to a present high of 65. I've checked the server status but I see no problems there. What are your thoughts on why the backlog is occurring?

Cheers,
Phil

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1695
Credit: 1042004384
RAC: 1404376

That is typical here

That is typical here Phil

When you start completing many more tasks than before the "Validation Pending" will climb as you have noticed.

It also depends on when the "wingman" finishes the same tasks.

Quite often when we are completing lots of tasks that "Validation Pending" will go above 100

You can check your pending tasks list and check the "Work unit click for details" to take a look at this.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109378039538
RAC: 35985209

RE: Is there a way to

Quote:
Is there a way to search by user ...


Unfortunately, no. You can obviously see things like total credit and RAC but (without keeping your own totals) there is no easy way to know how many tasks you have completed since you started, particularly if you go through a series of host IDs. In the BOINC directory, there is a file whose name starts 'job_log_einstein....' which is a continuing record of what a particular machine has done. It can easily be lost (eg if you wipe things and reinstall or have a disk failure) unless you keep it backed up. If it gets deleted, BOINC will recreate it but the previous history will be lost. If you wanted to produce some sort of record of your production over time, you could write a script to parse the information in that file and produce nicely formatted output.

Quote:
I've noticed that in the past 5 days my "Validation Pending" has tripled from the low 20s to a present high of 65. I've checked the server status but I see no problems there. What are your thoughts on why the backlog is occurring?


If you keep racing along and returning results so rapidly, you are bound to acquire an increasing number of 'pendings' at least until some sort of plateau may be reached :-).

As MAGIC has mentioned, 'pendings' are a necessary fact of life. I currently have 807 BRP6 pendings and around 1,600 overall. The number is always fluctuating, depending on the timeliness with which the companion result is returned. If you are unlucky enough to draw a 'wingman' that never returns the completed result and the replacement task (2 weeks later) also doesn't get returned, you can end up waiting a looooong time for the quorum to ultimately be completed. For this reason, if you ever decide to give up crunching, the kindest thing to do is to abort all onboard tasks and return them before uninstalling BOINC. The same goes for taking a break, say for a holiday. If you are switching off the machine for a week or two, abort and return the current tasks before you switch off. That way they can immediately be reissued to another computer that may return them in a more timely manner.

Cheers,
Gary.

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7022984931
RAC: 1834456

RE: I've noticed that in

Quote:
I've noticed that in the past 5 days my "Validation Pending" has tripled from the low 20s to a present high of 65. I've checked the server status but I see no problems there. What are your thoughts on why the backlog is occurring?


That one is simple in your case. You followed our advice to reduce the length of your prefetch queue.

While the behavior of your quorum partners--hence luck of the draw--is a major factor, it is also true that the older a task is at the moment yours is returned, the higher the chance is that you have outwaited your quorum partner by the time your return. If you outwaited him, and if your two results agree, then you get instant credit. Else the job goes to the pending state. On the plus side, in the last few days your partners have spent less time waiting for you.

RAC is a form of exponentially-weighted moving average, where the decay factor is elapsed-time based rather than the classic simply event-based. In the long run it will settle toward your average daily useful output, but the long run is weeks, not minutes, hours, or days.

When I want to keep close tabs on my actual production, I use a spreadsheet and add in an estimate of the value of my pending work to my reported total credit. Here on Einstein at the moment BRP6 jobs are almost always worth 4400 credits if accepted, so this works quite well. At other times and places the worth of pending work has been less predictable.

The fact your pending has climbed so much at a time you were getting daily credit totals so strong is a good indication that your current configuration is actually producing well above your current reported RAC, which has been climbing smartly.

BoincStats is a widely used credit reporting site.
You can look at their results for Einstein work by Gary Roberts starting here
and their display for you is here
The graphs tab may be particularly enlightening.

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7022984931
RAC: 1834456

RE: Here on Einstein at the

Quote:
Here on Einstein at the moment BRP6 jobs are almost always worth 4400 credits if accepted, so this works quite well.


I neglected to mention that part of your pending is CPU work of the "Gravitational Wave search O1 all-sky F v1.04" flavor. These are pretty uniformly worth 1000 credits. Just now you had 24 pending CPU units and 38 pending GPU units. Probably your pending CPU work will climb a good bit more, as until very recently you were still returning CPU work downloaded in a giant gulp around May 15, so your quorum partners had the better part of two weeks to get their answer in before you. Your most recently returned CPU work has only a day or so of aging, so naturally fewer of your partners will have beat you to the punch.

Florida Rancher
Florida Rancher
Joined: 4 Oct 13
Posts: 31
Credit: 23998436
RAC: 0

Great stuff everyone. I'm

Great stuff everyone. I'm still wondering how each of us knows how many, if any, pulsars we've discovered.

On different subject my Boinc "Detailed Stats" page seems a little off. I think these stats get updated sometime in the morning around 10 AM EDT or UTC-5. What's being reported and what I'm manually calculating is quite different.

Today's credit per day is 90,600 on the "Detailed Stats" page. But, at 8 PM UTC-5 time I write down both my "Work Done" and "Avg. Work Done" from the Boinc program. Last nights work done was 3,805,572 and the average work done was 110,526.

So let's speculate that the "Detailed Stats" page updates at 10 AM UTC-5 which would be 14 hours earlier than when I wrote down my 5/31 stats. According to the Boinc Program my last 24 hour period between 8 PM UTC-5 on 5/31/2016 & 8 PM UTC-5 on 6/1/2016 produced credit totaling 172,400. The previous 24 hour period produced 133,586 of credit.

If the "Detailed Stats" reporting is a 24 hour period beginning and ending at 10 AM UTC-5 my credit per day being reported for that time period is only 90,600. I understand that these two time periods differ by 14 hours but it is still difficult to reconcile a credit per day difference of 81,800.

If I add the two days together my written credit totals 305,986 while "Detailed Stats" totals only 210,293.

The only theory I can come up with is that the Boinc program gives you full credit when it updates while the Boinc "Detailed Stats" page takes into account pending validations. What are your thoughts on this.

Phil

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7022984931
RAC: 1834456

RE: The only theory I can

Quote:
The only theory I can come up with is that the Boinc program gives you full credit when it updates while the Boinc "Detailed Stats" page takes into account pending validations. What are your thoughts on this.


I don't know what specific data sources you are referring to.

It may be useful to note that the non-Einstein BOINC data sites rely on files which are published by Einstein and other BOINC projects to a standardized format. They are not continuously updated. Sometimes they are later than others, and not always does an external site catch the same release.

As your reporting is not at all uniform through the day, and the availability of already completed quorum partner results is not at a constant rate, it is quite natural that frequent sampling of credit awarded fluctuates quite widely in rate. This is part of the reason that RAC is calculated and provided, though it has properties that some people dislike so sharply as to declare it useless.

I don't think any of the major BOINC statistics sources attempt to credit you with pending work. While the credit here at Einstein is very predictable on some applications (including both that you run currently) it is wildly unpredictable on other applications around the BOINC world.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.