Welcome to MilkyWay@home

What is the meaning of "Consecutive valid tasks" statistic?

Message boards : Number crunching : What is the meaning of "Consecutive valid tasks" statistic?
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · Next

AuthorMessage
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 578
Credit: 18,845,239
RAC: 856
Message 49355 - Posted: 16 Jun 2011, 9:38:53 UTC

Well, I was following the numbers as I was tweeking the app_info file and it seems to be as I thought before:

1. The number of consecutive valid tasks is "Max tasks per day" - 10000.*

2. Anything that becomes inconclusive and than resend counts as a "consecutive valid task" in addition to increasing the "Max tasks per day" by one as well (if it's valid). That does not apply to tasks, that are resended because of some error and validate for the first error free returned result, i.e. there needs to be at least 2 valid results/WU for to be counted as "consecutive valid task". So this number was increasing pretty slow for me, but was never decreased or reseted to 0, all returned results were also valid so far.


* there is a little "but", which appeared yesterday at the end of the day, when I was not looking at every task so carefully anymore, just followed the numbers on the application page every few tasks:

Number of tasks completed:    130
Max tasks per day:            10131


Where is the additional task per day coming from?
ID: 49355 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
TJ

Send message
Joined: 12 Aug 09
Posts: 262
Credit: 92,631,041
RAC: 0
Message 49356 - Posted: 16 Jun 2011, 10:18:31 UTC
Last modified: 16 Jun 2011, 10:18:54 UTC

This is my part from: MilkyWay@Home 0.82 windows_x86_64 (ati14) at 10:13UTC
    Number of tasks completed 556
    Max tasks per day 10548
    Number of tasks today 263
    Consecutive valid tasks 4
    Average processing rate 289.5943528425
    Average turnaround time 0.03 days


So I think the max tasks per day grows every day.
And I don't believe that there where only 4 tasks in a row that where valid.
It is hard to say as they clear out fast from the database, but with a lot of errorr, credit could not be so high.

It would be very helpfull if Travis is going to explane this.


Greetings from,
TJ
ID: 49356 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 578
Credit: 18,845,239
RAC: 856
Message 49379 - Posted: 16 Jun 2011, 20:21:22 UTC

I finally had my first computation error, caused probably by a VPU recover... whatever that is.

So what changed on the application page...

Number of tasks completed     194
Max tasks per day             10194
Number of tasks today         90
Consecutive valid tasks       5


What I don't know (nothing suspicious in BOINC log) is, if both tasks running at that moment were trashed (I run two at once), but as the "Max tasks per day" was decreased by 1 probably only one task was affected, unless I got an additional task per day again... And "Consecutive valid tasks" was obviously reseted to 0.

So the summary of my observations would be: after running more than just few tasks, it's impossible to know the real number of consecutive valid tasks, specially after some invalid results. The "Max tasks per day - 10000" theory fails, because sometimes you get some bonus task out of nowhere in there and when you produce an error the number is decreased by one and not reseted to 10000. And under "consecutive valid tasks" are counted only those, that has got at least two valid results.
ID: 49379 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 578
Credit: 18,845,239
RAC: 856
Message 49697 - Posted: 28 Jun 2011, 14:36:02 UTC - in response to Message 49379.  

So, now that we have get the opportunity to produce lots of errors, the following seem not to be correct, at least not for consecutive errors:

The "Max tasks per day - 10000" theory fails, because sometimes you get some bonus task out of nowhere in there and when you produce an error the number is decreased by one and not reseted to 10000. And under "consecutive valid tasks" are counted only those, that has got at least two valid results.


Apparently the "Max tasks per day" value is reseted to 10000 after more than 1 consecutive errors. I've reported 7 results at once, so I can't say exactly, how many are needed, but for just 1 error the previous assumption should still be correct unless they have changed something.
ID: 49697 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
TJ

Send message
Joined: 12 Aug 09
Posts: 262
Credit: 92,631,041
RAC: 0
Message 49720 - Posted: 28 Jun 2011, 21:13:56 UTC - in response to Message 49697.  
Last modified: 28 Jun 2011, 21:14:23 UTC

Well I have looked at it for days as well and I have even used two systems with no good ati-cards, so they must use CPU.
But errors or not at no pc the Maximum tasks per day was never reset to 10000.
It is counting upwards, even when consecutive tasks drops to zero again.
Really prof. Travis has to give us the answer of this use.
Greetings from,
TJ
ID: 49720 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ExtraTerrestrial Apes
Avatar

Send message
Joined: 1 Sep 08
Posts: 204
Credit: 219,354,537
RAC: 0
Message 49921 - Posted: 5 Jul 2011, 20:47:36 UTC - in response to Message 49379.  

Well, just don't bother with the "max tasks per day". As long as they stay at approximately 10k their value doesn't matter.

What does matter is the definition of "Consecutive valid tasks", especially since with insta-purge it is about the best feedback on result reliabilty which we have so far. I think we've got the meaning sorted out pretty well - this metric measures just what the name implies.

However, just looking at the recent number of valid WUs since the last error is highly random, if you get some errors at all. I.e. you have to watch this value carefully to get an idea of what's going on.

Suggestion: can we get this value averaged over something between 1 day and 7 days? It should be rather easy for the server: everytime "Consecutive valid tasks" is reset to zero it can store internally:

- time / date
- number of consecutive valid tasks before reset
- drop old values

From this one can easily deduce the average error rate as
1 - sum(consecutive valid tasks) / sum(consecutive valid tasks + 1)
That would be very helpful to diagnose problems due to OC despite the presence of insta-purge.

I finally had my first computation error, caused probably by a VPU recover... whatever that is.

Your GPU (VPU) chocked badly upon an error an stopped responding. The driver noticed this and reset the GPU, saving you from a freeze or blue screen.

MrS
Scanning for our furry friends since Jan 2002
ID: 49921 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 578
Credit: 18,845,239
RAC: 856
Message 49924 - Posted: 5 Jul 2011, 22:22:59 UTC - in response to Message 49921.  

What does matter is the definition of "Consecutive valid tasks", especially since with insta-purge it is about the best feedback on result reliabilty which we have so far. I think we've got the meaning sorted out pretty well - this metric measures just what the name implies.

Well, the problem is that "consecutive valid tasks" is just not what the name implies, actually it's "consecutive valid tasks with more than one valid result returned". And most of the WUs here are validated after just one.

And that makes this calculation wrong:
From this one can easily deduce the average error rate as
1 - sum(consecutive valid tasks) / sum(consecutive valid tasks + 1)
That would be very helpful to diagnose problems due to OC despite the presence of insta-purge.

Since my last post in this thread my value for "consecutive valid tasks" was not reseted once, I was watching it, it's ATM at 172 and I'm doing 80-100 WUs a day. So for this calculation should be right, my value should have been after this 7 days somewhere around 560-700. But it's not, not even close to that.
ID: 49924 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
TJ

Send message
Joined: 12 Aug 09
Posts: 262
Credit: 92,631,041
RAC: 0
Message 49925 - Posted: 5 Jul 2011, 22:35:19 UTC - in response to Message 49924.  

Since my last post in this thread my value for "consecutive valid tasks" was not reseted once, I was watching it, it's ATM at 172 and I'm doing 80-100 WUs a day. So for this calculation should be right, my value should have been after this 7 days somewhere around 560-700. But it's not, not even close to that.

Link have you watched it 24 hours continious? Past week I could'nt sleep and watched it for 2.5 hours on two screens. One with the BOINC manager and the other with the tasks list and presed F5 every time 2 tasks where sent from the BOINC tray. And found no error. The CVT (consecutive valid tasks) did not increase with every validated task.
Today I found a first validate error and the CVT was reset to 1. (The max tasks per day not to 10000).
Now half an hour later several tasks are validated (credit has increased) but the CVT is still 1.
Have you an idea?
Greetings from,
TJ
ID: 49925 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ba2pt

Send message
Joined: 9 May 11
Posts: 15
Credit: 22,575,796
RAC: 0
Message 49932 - Posted: 6 Jul 2011, 4:04:41 UTC - in response to Message 49925.  
Last modified: 6 Jul 2011, 4:05:05 UTC

Since my last post in this thread my value for "consecutive valid tasks" was not reseted once, I was watching it, it's ATM at 172 and I'm doing 80-100 WUs a day. So for this calculation should be right, my value should have been after this 7 days somewhere around 560-700. But it's not, not even close to that.

Link have you watched it 24 hours continious? Past week I could'nt sleep and watched it for 2.5 hours on two screens. One with the BOINC manager and the other with the tasks list and presed F5 every time 2 tasks where sent from the BOINC tray. And found no error. The CVT (consecutive valid tasks) did not increase with every validated task.
Today I found a first validate error and the CVT was reset to 1. (The max tasks per day not to 10000).
Now half an hour later several tasks are validated (credit has increased) but the CVT is still 1.
Have you an idea?


Sounds like both you and Link are seeing the same thing: despite initial expectations to the contrary, there's not a 1:1 relationship between the number of valid tasks submitted and the CVT count. Without more specificity, all we can call it is a loose measure of reliability. If your CVT is resetting all the time, then you may have problems with system stability. But any conclusions beyond that are going to be unjustified.
ID: 49932 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 578
Credit: 18,845,239
RAC: 856
Message 49937 - Posted: 6 Jul 2011, 7:14:51 UTC - in response to Message 49925.  
Last modified: 6 Jul 2011, 7:16:26 UTC

Since my last post in this thread my value for "consecutive valid tasks" was not reseted once, I was watching it, it's ATM at 172 and I'm doing 80-100 WUs a day. So for this calculation should be right, my value should have been after this 7 days somewhere around 560-700. But it's not, not even close to that.

Link have you watched it 24 hours continious? Past week I could'nt sleep and watched it for 2.5 hours on two screens. One with the BOINC manager and the other with the tasks list and presed F5 every time 2 tasks where sent from the BOINC tray. And found no error. The CVT (consecutive valid tasks) did not increase with every validated task.

Yes, didn't miss anything, now with 170+ "consecutive valid tasks" I even don't need to watch so exactly anymore, I need about two days to make that many WUs on my HD3850.

Next time you can't sleep, watch out for WUs with more than one valid result, i.e. those that become inconclusive at first and get resend (not computing or validate errors, they don't count). You'll probably come to the same result as me: they are the ones, that increase the CVT value.
ID: 49937 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Sunny129
Avatar

Send message
Joined: 25 Jan 11
Posts: 271
Credit: 346,072,284
RAC: 0
Message 49944 - Posted: 6 Jul 2011, 13:46:35 UTC - in response to Message 49924.  
Last modified: 6 Jul 2011, 13:47:21 UTC

Well, the problem is that "consecutive valid tasks" is just not what the name implies, actually it's "consecutive valid tasks with more than one valid result returned".

well, in all fairness, there is no such thing as a solitary valid task - that is, you might be the first to successfully return a result without error, but it will not get validated until a 2nd successful result is returned by another host. only then will your error-free result get counted as "valid," and only if the successful results are "similar."

that being said, you do make a good point, and that is that the one's CVT statistic can be quite dependent on results of other hosts. for example, i might have the only successful result awaiting validation for a WU that's only one result away from reaching the maximum # of errors allowed for that particular WU b/c 4 other hosts have already returned errors. if i'm not the only one with a "task in progress" for this particular WU, and if another host finishes and returns its result before my host does, and if that result is an error, then the WU reaches the maximum # of allowed errors, and nobody ends up receiving credit for that particular WU. even if my result were to end up being error-free, it would never get validated b/c the server received the max # of errors allowed for that WU before it reached the minimum quorum of 2 validated results.

another example i can think of is one in which the max # of error/total/success tasks has not yet been reached, and yet more than one successful result has been returned but not validated. suppose that my successful result has been reported to the server along with one other successful result from host B. if they haven't validated, then that means that the results were not similar enough. when the third successful result gets reported by another host, 3 things could happen: 1) it might not get validated b/c it is not similar enough to either my result or host B's result, resulting in 3 tasks whose validations are now inconclusive 2) it could be similar enough to my result, thus validating that result and mine, but invalidating host B's result, or 3) it could be similar enough to host B's result, thus validating that result along with host B's result, and invalidating my result, even though it was error-free.


just some food for thought...
ID: 49944 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 578
Credit: 18,845,239
RAC: 856
Message 49946 - Posted: 6 Jul 2011, 15:17:28 UTC - in response to Message 49944.  

you might be the first to successfully return a result without error, but it will not get validated until a 2nd successful result is returned by another host. only then will your error-free result get counted as "valid," and only if the successful results are "similar."

The most of results here are validated after only one successful result, like this one for example:



Only when needed a 2nd (or more) results are send out, i.e. when they become inconclusive with itself (?), whatever causes that. If at least two results were needed for validation in general, there would be two send out at once (like Seti is doing it for example) and not after the first one became inconclusive. Just look at *_0 tasks, you'll see that most of them are send only to your host.
ID: 49946 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Sunny129
Avatar

Send message
Joined: 25 Jan 11
Posts: 271
Credit: 346,072,284
RAC: 0
Message 49947 - Posted: 6 Jul 2011, 15:44:32 UTC

oops...i just realized that i was actually referencing S@H WU's for the above examples, and not MW@H WU's. so i stand corrected on that, as well as the fact that some MW@H have a minimum quorum of 1. that being said, there are MW@H WU's which have a minimum quorum of 2, and therefore require more than 1 successful result. now, which type of WU is more common - those with a minimum quorum of 1 or those with a minimum quorum of 2 - i have no idea. regardless, b/c such MW@H WU's do exist, one's CVT statistic can still be influenced/affected by the results of another host. and b/c of this, we still can't get a very clear picture of what percentage of CVT resets are caused by our own errors, and what percentage of CVT resets are caused by something else (whether it be the errors of another host, a minimum quorum issue, etc).
ID: 49947 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
TJ

Send message
Joined: 12 Aug 09
Posts: 262
Credit: 92,631,041
RAC: 0
Message 49956 - Posted: 6 Jul 2011, 19:30:09 UTC

Well it is still complicated.

I can live with the idea of Link, but what Sunny129 syas is reasonable as well.
But if a WU is sent to 1, 2 or 3 clients, when I see that I got credit for a WU then that WU was error free. Right? I don't need to look if this WU was sent to others as well as it was granted credit.

As I mentioned a few post earlier, I saw that my CVT was set to 1 and I got credit for at least 24 WU's the CVT was still 1. The maximum allowed tasks was increased little, but that amount can never be reached on my system even the 10000 not.

So to be honest I still don't understand the CVT.
Greetings from,
TJ
ID: 49956 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
cncguru
Avatar

Send message
Joined: 11 Jun 10
Posts: 329
Credit: 1,166,222,661
RAC: 0
Message 49961 - Posted: 7 Jul 2011, 0:27:16 UTC
Last modified: 7 Jul 2011, 0:31:06 UTC

MilkyWay@Home (anonymous platform, ATI GPU)
Number of tasks completed 152759
Max tasks per day 14614
Number of tasks today 3735
Consecutive valid tasks 1003
Average processing rate 290.0261040405
Average turnaround time 0.01 days

This guy is running (4) HD 6970's @ 950 core 500 memory
ID: 49961 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
cncguru
Avatar

Send message
Joined: 11 Jun 10
Posts: 329
Credit: 1,166,222,661
RAC: 0
Message 49966 - Posted: 7 Jul 2011, 5:29:04 UTC

Energizer Bunny!!!

MilkyWay@Home (anonymous platform, ATI GPU)
Number of tasks completed 153649
Max tasks per day 15519
Number of tasks today 256
Consecutive valid tasks 1237
Average processing rate 294.77774899042
Average turnaround time 0.01 days
ID: 49966 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
cncguru
Avatar

Send message
Joined: 11 Jun 10
Posts: 329
Credit: 1,166,222,661
RAC: 0
Message 49971 - Posted: 7 Jul 2011, 9:45:08 UTC

MilkyWay@Home (anonymous platform, ATI GPU)
Number of tasks completed 154389
Max tasks per day 16266
Number of tasks today 1002
Consecutive valid tasks 1404
Average processing rate 291.83655778383
Average turnaround time 0.01 days

I have to admit I never paid any attention to these stats before this thread appeared. I have two others crunching Milkyway ATM and neither of them have gotten anywhere near 1400 CVT yet they are still accumulating CS just like this one.
So yeah, What is the meaning of "Consecutive valid tasks" statistic?
ID: 49971 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
cncguru
Avatar

Send message
Joined: 11 Jun 10
Posts: 329
Credit: 1,166,222,661
RAC: 0
Message 49981 - Posted: 7 Jul 2011, 15:02:56 UTC

MilkyWay@Home (anonymous platform, ATI GPU)
Number of tasks completed 155372
Max tasks per day 17258
Number of tasks today 1999
Consecutive valid tasks 1650
Average processing rate 277.33980317894
Average turnaround time 0.01 days

ID: 49981 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
cncguru
Avatar

Send message
Joined: 11 Jun 10
Posts: 329
Credit: 1,166,222,661
RAC: 0
Message 49991 - Posted: 7 Jul 2011, 21:18:50 UTC

MilkyWay@Home (anonymous platform, ATI GPU)
Number of tasks completed 156484
Max tasks per day 18393
Number of tasks today 3150
Consecutive valid tasks 1920
Average processing rate 291.13470624673
Average turnaround time 0.01 days


This stat is meaningless unless someone can show me a setting to tweak that will actually directly effect it.
I can't say, "Hey, can someone match this?" because I sure didn't change anything to try to get this to happen, it's just an anomaly.
ID: 49991 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
cncguru
Avatar

Send message
Joined: 11 Jun 10
Posts: 329
Credit: 1,166,222,661
RAC: 0
Message 50005 - Posted: 8 Jul 2011, 14:34:39 UTC

MilkyWay@Home (anonymous platform, ATI GPU)
Number of tasks completed 159240
Max tasks per day 21193
Number of tasks today 1881
Consecutive valid tasks 2567
Average processing rate 285.44982941387
Average turnaround time 0.01 days
ID: 50005 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · Next

Message boards : Number crunching : What is the meaning of "Consecutive valid tasks" statistic?

©2024 Astroinformatics Group