A BIG increase in VALIDATION time

Denis Puhar, dr. med.
Denis Puhar, dr...
Joined: 5 Nov 09
Posts: 36
Credit: 7,006,583
RAC: 0
Topic 196767

Hi!

I decided a few weeks ago to crunch as many GPU Arecibo tasks as possible and with doing so, devoting a larger proportion of my TOTAL computing power (which is rather small) to E@H.

Yet, I've noticed, that the increase in my average credit and especially BS-RAC is not proportional and very small in comparison to how many WUs I complete in one day (or even 2 or 3 weeks).

For example:

Tonight after midnight (CET, GMT + 1) I got 196 new WUs and still 8 to complete from my previous 'package' of tasks.

In 17h of computing I've completed 37 WUs (29 'new' and the 8 'old'), yet as I checked at around 17.45pm, CET, there were 44 finished WUs (ALL without errors or 'inconclusive' tasks) on PENDING (some still waiting from 11.,12. or 13. of January).

As a consequence my average (especially BS-RAC) goes up very slowly, almost as an 'exponential' function of the number of finished WUs (X axis) to the ratio of those which are pending (Y axis).

To some degree this is perfectly logical but I find it a bit frustrating, that sometimes there are more WUs on PENDING than those that I finish and I'm granted credit for.

D

“A little knowledge is a dangerous thing. So is a lot.” - Albert EINSTEIN

Patrick
Patrick
Joined: 2 Aug 12
Posts: 70
Credit: 2,358,155
RAC: 0

A BIG increase in VALIDATION time

If you crunch 44 WU´s in 24 hours your rac would be 44*500credits for brp so if you compute this every day your rac will be 22000 credits but it takes a while until your wu are crunched from other users or are validated so in 1 or 2 weeks your rac should be at around 22000.
I don´t know where your problem is.
If you can´t wait and know how many wu´s you compute you can calculate the rac for yourself.

Patrick
Patrick
Joined: 2 Aug 12
Posts: 70
Credit: 2,358,155
RAC: 0

If i watch at your gpu

If i watch at your gpu runtimes and divide 86400seconds which is 1 day through 1630 seconds of your gpu time you will finish 53 wu´s a day.That´s a RAC of 26500 if you crunch with the standard utlization factor of 1.0(1 Task).
With your card you could crunch 2 perhaps 3 simultaneously if you want that and if its profitable from the sight of the runtimes.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,206
Credit: 43,314,748,984
RAC: 44,616,735

RE: Tonight after midnight

Quote:
Tonight after midnight (CET, GMT + 1) I got 196 new WUs and still 8 to complete from my previous 'package' of tasks.


Does it seem a little strange to you that BOINC doesn't get any new work until your work cache is almost completely exhausted? Just 8 tasks left from the previous batch. And then you get almost 200 new tasks in one big hit. Seems a little sub-optimal, doesn't it?

The reason for this is that you are using a V7 BOINC where the meanings of the two preference settings that control work cache size are completely reversed from what they used to mean in V6 and earlier BOINCs. Unfortunately, E@H uses a highly customised version of older server code so when you look at your prefs on E@H, you see things like
[pre]
Computer is connected to the Internet about every
Leave blank or 0 if always connected.
BOINC will try to maintain at least this much work (max 10 days). 0.01 days
Maintain enough work for an additional 4 days
[/pre]

So, if you are not aware of the V7 changes, you might very well have the first setting at zero or something very close to it - like the 0.01 days shown above.

Under V7, the first setting is your basic cache size, let's call it the low water mark. If you want (say) 4 days, that's exactly what you should set there. The second setting is an extra setting on top of the base setting. This defines a high water mark (base setting + extra days). So, in the above example the base cache size is 0.01 days and when work is requested it will top up to a 4.01 day high water mark. This appears to be what is happening to you.

Under normal circumstances you would set the extra days to something close to zero. If the project were to announce a potential outage of say 6 days maximum, you could set this second setting high enough (4 day base cache plus say 3 extra days) to give a high water mark beyond the period of the outage. Once the extra work had downloaded you could set it back to zero as you would now have enough work to outlast the outage event.

Quote:
In 17h of computing I've completed 37 WUs (29 'new' and the 8 'old'), yet as I checked at around 17.45pm, CET, there were 44 finished WUs (ALL without errors or 'inconclusive' tasks) on PENDING (some still waiting from 11.,12. or 13. of January).


Think about what you've just described. You've just returned 29 'new' tasks as you call them, a few short hours after you received them. There would be something a bit strange if the majority of these were NOT pending. Many of your crunching partners will have caches longer than 17 hours. They won't have even started the tasks by the time you have returned them.

On top of that, there is always a significant proportion of all tasks sent out that error out or are never returned by the first recipient. I don't know the fraction but it may be as much as 20% or more. So, to have pendings in your list that are older then 14 days is not unusual. You will always be able to find examples even up to a month or more after the original task was distributed. Everybody is affected by this so there is not much point complaining about it.

There are things you can do to minimise the level of pendings. You could set your preferences to maintain a steady cache size rather than one which runs down to zero before being replenished. If you maintain an X day cache size, every new task you start will have already been on your system for X days and that means that many of your crunching partners may have already returned their copy. You should not set X too high. if you do, you are subjecting your crunching partners to the same annoyance you are complaining about. 3 to 4 days would be a very suitable cache size for this project. Many people would be setting less than this. If you were supporting multiple projects, a lower value would probably be more appropriate.

Another thing to realise is that RAC is a long term moving average and it probably takes about a month after a significant change for the new level to be reached. Also, the level of pendings will eventually cease rising. There will come the time when every single new pending created will be offset by an old pending finally being completed. A 'steady state' will be reached.

Cheers,
Gary.

Nobody316
Nobody316
Joined: 14 Jan 13
Posts: 141
Credit: 2,008,126
RAC: 0

RE: To some degree this is

Quote:
To some degree this is perfectly logical but I find it a bit frustrating, that sometimes there are more WUs on PENDING than those that I finish and I'm granted credit for.

Quote:
On top of that, there is always a significant proportion of all tasks sent out that error out or are never returned by the first recipient. I don't know the fraction but it may be as much as 20% or more. So, to have pendings in your list that are older then 14 days is not unusual. You will always be able to find examples even up to a month or more after the original task was distributed. Everybody is affected by this so there is not much point complaining about it.

Just the other day I had almost 10k worth of Pending credit. I didn't count how many there was. I just started back working after afew days out. I could careless about credit/points or BS-RAC. I only worry about failed units because I am here to help with researching. But this is just me I think having 1 billion points/credits is worthless.... Help finding ET or new star or pulsar or black hole or anything else is priceless. If everything holds out with in the next week or 2 I am sure my pending tasks will double but that is fine as long as it gets done in the end.

Just thoughts of Nobody316

PC setup MSI-970A-G46 AMD FX-8350 8 core OC'd 4.45GHz 16GB ram PC3-10700 Geforce GTX 650Ti Windows 7 x64 Einstein@Home

juan BFP
juan BFP
Joined: 18 Nov 11
Posts: 839
Credit: 421,443,712
RAC: 0

I have more than 1370 WU

I have more than 1370 WU pending for credits, at 500 each = > 600K in pendings more than my hosts do in a compleate 1 day of work.

lHj2ixL.jpg

 

Comment viewing options

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